Eta calculation method, device and equipment for cross-border whole vehicle business and medium

Abstract

The application relates to the technical field of cross-border logistics transportation management, and provides an ETA calculation method, device, equipment and medium for cross-border whole vehicle business, wherein the method is used for calculating congestion delay and customs clearance delay by acquiring real-time traffic data and customs clearance time prediction data; the basic transportation time is calculated based on the acquired real-time traffic data and customs clearance time prediction data, and the congestion delay and the customs clearance delay are added to generate initial path transportation time; driver historical driving data is acquired, a personalized correction coefficient library is constructed based on the historical driving data, the initial path transportation time is corrected according to the personalized correction coefficient library, and the corrected transportation time is obtained; multi-dimensional parameter data is acquired, the multi-dimensional parameters are input into an intelligent time limit prediction model trained based on historical transportation data, and initial ETA is generated; node reported data and traffic congestion data are acquired in real time during transportation, the initial ETA is dynamically corrected, and the final corrected ETA is output.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of cross-border logistics transportation management, and particularly relates to an ETA calculation method, device and equipment for cross-border whole vehicle business and a medium. BACKGROUND

[0002] In the cross-border whole vehicle transportation business, accurately predicting the estimated time of arrival (ETA) of goods is a core requirement for ensuring logistics efficiency and customer experience. Traditional ETA calculation methods generally have the following technical defects:

[0003] 1. Single data dimension: Existing solutions are mostly based on static distance or historical average algorithms, and only simply combine real-time traffic of map navigation, without integrating delay prediction of customs clearance links (such as customs declaration document pre-audit status, port congestion degree, inspection probability, etc.), resulting in that the time efficiency uncertainty of the customs clearance link cannot be quantitatively evaluated.

[0004] 2. Ignoring individual differences of drivers: Traditional methods do not establish a correlation model between driving behavior of drivers and transportation time efficiency, and cannot make individualized correction on ETA according to driving habits (such as average speed, frequency of sudden acceleration / braking, etc.) of different drivers, resulting in that the ETA prediction results of different drivers on the same route lack accuracy.

[0005] 3. Insufficient fusion of multi-dimensional factors: Existing technologies do not construct an intelligent prediction model containing multi-dimensional parameters such as vehicle type, route characteristics, weather, holidays, port traffic, driver score, etc., and it is difficult to dynamically respond to time efficiency influencing factors in complex scenarios, and the ETA error often exceeds 1 hour, which cannot meet the high requirements of cross-border logistics on time efficiency accuracy.

[0006] 4. Lack of dynamic correction mechanism: There is a lack of dynamic calibration of ETA by real-time node data (such as reporting at key positions, sudden traffic incidents) in the transportation process, resulting in that the prediction results cannot be adjusted in real time according to the actual transportation state, further exacerbating the lag of time efficiency prediction.

[0007] At present, there is no solution in the existing technology that organically combines real-time traffic data, customs clearance prediction model, driver behavior characteristics and multi-dimensional influencing factors, and realizes dynamic and accurate calculation of ETA through intelligent algorithms. Therefore, how to break through the limitations of single data dimension and static rules, and construct an ETA calculation method of multi-source data fusion, individualized correction and dynamic adjustment, has become a technical problem to be solved in the field. SUMMARY

[0008] The present application provides an ETA calculation method, device and equipment for cross-border whole vehicle business and a medium, aiming to solve the problem that there is no solution in the existing technology that organically combines real-time traffic data, customs clearance prediction model, driver behavior characteristics and multi-dimensional influencing factors, and realizes dynamic and accurate calculation of ETA through intelligent algorithms.

[0009] In a first aspect, the application provides an ETA calculation method for cross-border whole vehicle business, comprising:

[0010] obtaining real-time traffic data and customs clearance time prediction data, wherein the real-time traffic data is obtained by calling a map application interface and is used to calculate congestion delay caused by road congestion; the customs clearance time prediction data is obtained by a customs declaration and inspection time prediction model and is used to predict clearance delay caused by customs declaration pre-audit, port congestion and inspection probability;

[0011] based on the obtained real-time traffic data and customs clearance time prediction data, combining line data of a map or a transportation system and average driving speed of a vehicle under normal conditions, calculating basic transportation time, adding congestion delay and clearance delay, and generating initial path transportation time;

[0012] obtaining driver historical driving data including average speed, brake frequency and sudden acceleration, constructing a personalized correction coefficient library based on the historical driving data, correcting the initial path transportation time according to the personalized correction coefficient library, and obtaining corrected transportation time;

[0013] obtaining multi-dimensional parameter data including vehicle type, line, weather, holiday, port flow and driver score, inputting the multi-dimensional parameter data into an intelligent time limit prediction model trained based on historical transportation data to generate an initial ETA, obtaining node reporting data and traffic congestion data in real time during transportation, dynamically correcting the initial ETA, and outputting a final corrected ETA.

[0014] In some embodiments, based on the obtained real-time traffic data and customs clearance time prediction data, combining line data of a map or a transportation system and average driving speed of a vehicle under normal conditions, calculating basic transportation time, comprises: obtaining total distance of a transportation line from a map or a transportation system, determining average driving speed of a vehicle under normal conditions based on historical operation data, vehicle type characteristics or industry experience value, dividing the total distance by the average driving speed to obtain basic driving time without considering traffic and clearance factors; the real-time traffic data is used to calculate driving time of each road section increased due to congestion as congestion delay, and the customs clearance time prediction data is used to determine clearance delay corresponding to customs declaration pre-audit status, port congestion degree and inspection probability.

[0015] In some embodiments, the driver historical driving data includes average speed, brake frequency, and sudden acceleration, and the personalized correction coefficient library is constructed based on the historical driving data, including: collecting the average speed, brake frequency per mile, and sudden acceleration times of the driver in the historical transportation task through the vehicle-mounted sensor or the driver terminal device, for statistical analysis according to a preset time period, establishing a corresponding relationship between different driving behavior characteristics and transportation time correction coefficients, the correction coefficient is determined by the ratio of the actual arrival time to the basic transportation time in the historical transportation task, and a personalized correction coefficient library for each driver is formed.

[0016] In some embodiments, the initial path transportation time is corrected according to the personalized correction coefficient library to obtain a corrected transportation time, including: according to the identity of the driver currently performing the transportation task, matching the corresponding driving behavior correction coefficient from the personalized correction coefficient library, multiplying the initial path transportation time by the correction coefficient, and if it is detected that the sudden acceleration frequency exceeds a preset threshold, continuous sudden braking, or other risky driving behaviors exist in the current driving behavior, adding a preset risk delay to the correction result to form a corrected transportation time considering the individual differences of the driver.

[0017] In some embodiments, the multi-dimensional parameters are input into an intelligent time efficiency prediction model trained based on historical transportation data to generate an initial ETA, including: using historical transportation data as a training set, using vehicle type, route, weather, holiday, port flow, and driver score as input features, and using actual arrival time as output label, and using a machine learning algorithm to train an intelligent time efficiency prediction model; the machine learning algorithm includes but is not limited to random forest, gradient boosting tree, or neural network algorithm; the multi-dimensional parameters of the current transportation task are input into the model, and the initial ETA considering multi-dimensional factors is calculated and output by the intelligent time efficiency prediction model.

[0018] In some embodiments, the initial ETA is dynamically corrected by real-time acquisition of node reporting data and traffic congestion data during transportation to output a final corrected ETA, including: real-time reporting of the current position of the vehicle and the time of arrival at the preset key node through the vehicle-mounted GPS device or the driver terminal, acquiring real-time updated traffic congestion data and port customs status change data, and recalculating the congestion delay of the current road section and the customs delay of the remaining road section based on the newly acquired traffic congestion data and port customs status change data; combining the remaining transportation distance and the real-time driving behavior data of the driver, iteratively updating the initial ETA, and outputting the final corrected ETA containing real-time state.

[0019] In some embodiments, the method further comprises: when the time difference between the initial ETA output by the intelligent time limit prediction model and the promised time limit exceeds a preset threshold, triggering a multi-end collaborative early warning mechanism, and pushing a notification containing real-time predicted time limit, delay risk factors and alternative solutions to the customer side; if the number of consecutive rapid accelerations or the frequency of rapid braking in the driver's driving behavior data exceeds the risk threshold, the system automatically sends a driving behavior warning to the driver's end and synchronously adjusts the risk delay in the corrected transportation time, forming a dynamic time limit correction mechanism including risk intervention.

[0020] In a second aspect, the application provides an ETA calculation device for cross-border whole vehicle business, comprising:

[0021] A data acquisition unit is configured to acquire real-time traffic data and customs clearance time prediction data, wherein the real-time traffic data is acquired by calling a map application program interface and is used to calculate congestion delay caused by road congestion; and the customs clearance time prediction data is obtained by a customs declaration and inspection time prediction model and is used to predict clearance delay caused by customs declaration pre-audit, port congestion and inspection ratio;

[0022] A time generation unit is configured to calculate a basic transportation time based on the acquired real-time traffic data and customs clearance time prediction data, in combination with line data of a map or transportation system and average driving speed of a vehicle under normal conditions, and to generate an initial path transportation time by adding congestion delay and clearance delay;

[0023] A time correction unit is configured to acquire driver historical driving data including average speed, brake frequency and rapid acceleration, to construct a personalized correction coefficient library based on the historical driving data, and to correct the initial path transportation time according to the personalized correction coefficient library to obtain a corrected transportation time;

[0024] A dynamic correction unit is configured to acquire multi-dimensional parameter data including vehicle model, line, weather, holiday, port flow and driver score, to input the multi-dimensional parameter data into an intelligent time limit prediction model trained based on historical transportation data to generate an initial ETA, and to dynamically correct the initial ETA by acquiring node reporting data and traffic congestion data in real time during transportation to output a final corrected ETA.

[0025] In a third aspect, the application further provides a computer device, comprising:

[0026] a memory and a processor;

[0027] the memory is configured to store a computer program;

[0028] the processor is configured to execute the computer program and implement the steps of the ETA calculation method for cross-border whole vehicle business according to the first aspect when executing the computer program.

[0029] In a fourth aspect, the present application also provides a computer readable storage medium storing a computer program, which, when executed by a processor, causes the processor to implement the steps of the ETA calculation method for cross-border whole vehicle business according to the first aspect.

[0030] The ETA calculation method, device, equipment and medium for cross-border whole vehicle business provided by the embodiments of the present application break through the limitation of traditional static rules by integrating real-time traffic data, customs clearance time prediction model, driver driving behavior data and multi-dimensional influence factors, control the ETA error within ± 15 minutes, and significantly improve the accuracy of time efficiency prediction. Based on the historical driving data of the driver, a personalized correction coefficient library is constructed, and the ETA is dynamically adjusted according to the driving habits of different drivers, solving the drawbacks of the traditional method of "one-size-fits-all" prediction, and improving the flexibility of transportation plan. By collecting real-time node reporting data and traffic congestion information, the ETA is dynamically corrected to ensure that the prediction result is updated in real time with the transportation state, effectively responding to the sudden delay risk in cross-border transportation. By introducing multi-dimensional parameters such as vehicle type, route, weather, holidays and the like to train an intelligent prediction model, the system can fully respond to the diversified time efficiency influence factors in cross-border transportation and improve the adaptability of the system to complex scenarios.

[0031] It should be understood that the foregoing general description and the following detailed description are only exemplary and explanatory, and cannot limit the present application. BRIEF DESCRIPTION OF DRAWINGS

[0032] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings needed in the embodiment description will be briefly introduced. Obviously, the drawings in the following description are some embodiments of the present application, and other drawings can also be obtained by those skilled in the art without creative labor.

[0033] Figure 1 is a step schematic flow chart of an ETA calculation method for cross-border whole vehicle business provided by an embodiment of the present application;

[0034] Figure 2 is a structural schematic diagram of an ETA calculation device for cross-border whole vehicle business provided by an embodiment of the present application;

[0035] Figure 3 is a structural schematic block diagram of a computer equipment provided by an embodiment of the present application.

[0036] It should be understood that the foregoing general description and the following detailed description are only exemplary and explanatory, and cannot limit the present application. DETAILED DESCRIPTION

[0037] The technical solutions in the embodiments of the present application will be described clearly and completely in combination with the drawings in the embodiments of the present application. Obviously, the described embodiments are some of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative work belong to the scope of protection of the present application.

[0038] The flowcharts shown in the drawings are only exemplary and do not necessarily include all contents and operations / steps, nor are they necessarily executed in the described order. For example, some operations / steps can be further decomposed, combined or partially merged, so that the actual execution order can be changed according to the actual situation.

[0039] It should be understood that, in order to clearly describe the technical solutions of the embodiments of the present application, in the embodiments of the present application, the terms "first", "second", etc. are used to distinguish the same items or similar items with basically the same functions and effects. Those skilled in the art can understand that the terms "first", "second", etc. do not limit the quantity and execution order, and the terms "first", "second", etc. do not necessarily mean different.

[0040] It should be understood that the terms used in the present application are only for the purpose of describing specific embodiments and do not intend to limit the present application. As used in the present application specification and the appended claims, unless otherwise clear from the context, the singular forms "a", "an" and "the" are intended to include the plural forms.

[0041] It should also be understood that the term "and / or" used in the present application specification and the appended claims means any combination of one or more of the associated listed items and all possible combinations, and includes these combinations.

[0042] Some embodiments of the present application will be described in detail below in combination with the drawings. The following embodiments and features in the embodiments can be combined with each other without conflict.

[0043] In cross-border whole vehicle transportation business, accurate prediction of the estimated time of arrival (ETA) of goods is the core requirement to ensure logistics efficiency and customer experience. The traditional ETA calculation method has the following technical defects:

[0044] 1. Single data dimension: existing solutions are mostly based on static distance or historical average algorithm, only combining the real-time traffic of map navigation, without integrating the delay prediction of customs clearance link (such as customs document pre-audit status, port congestion degree, inspection probability, etc.), resulting in the uncertainty of the time efficiency of the customs clearance link cannot be quantitatively evaluated.

[0045] 2. Ignoring individual differences of drivers: The traditional method does not establish a correlation model between the driving behavior of the driver and the transportation time limit, and cannot make individualized corrections to the ETA according to the driving habits of different drivers (such as average speed, frequency of sudden acceleration / braking, etc.), resulting in a lack of accuracy in the ETA prediction results of different drivers on the same route.

[0046] 3. Insufficient fusion of multi-dimensional factors: The existing technology does not construct an intelligent prediction model containing multi-dimensional parameters such as vehicle type, route characteristics, weather, holidays, port flow, and driver rating, making it difficult to dynamically respond to time limit influencing factors in complex scenarios, and the ETA error often exceeds 1 hour, which cannot meet the high requirements of cross-border logistics on time limit accuracy.

[0047] 4. Lack of dynamic correction mechanism: There is a lack of dynamic calibration of ETA with real-time node data (such as reports from key locations and sudden traffic incidents) during transportation, resulting in a lack of real-time adjustment of prediction results according to actual transportation conditions, further exacerbating the lag of time limit prediction.

[0048] Currently, there is no solution in the existing technology that combines real-time traffic data, customs clearance prediction model, driver behavior characteristics, and multi-dimensional influencing factors, and realizes dynamic and accurate calculation of ETA through intelligent algorithms. Therefore, how to break through the limitations of single data dimension and static rules, and construct an ETA calculation method with multi-source data fusion, individualized correction, and dynamic adjustment, has become a technical problem to be solved in the field.

[0049] Please refer to Figure 1 , Figure 1 is a schematic flowchart of the ETA calculation method for cross-border whole vehicle business provided by an embodiment of the present application. The ETA calculation method for cross-border whole vehicle business can be realized by a computer device, which can be deployed on a single server or a server cluster. It can also be deployed on a handheld terminal, a notebook computer, a wearable device, or a robot, etc.

[0050] It should be noted that the acquisition of any information involved in the provided method is in accordance with relevant regulations and with the consent of the user, and does not infringe on the privacy of the user or violate relevant laws and regulations.

[0051] Specifically, as Figure 1 shown, the ETA calculation method for cross-border whole vehicle business provided includes steps S101 to S104, which are described in detail as follows:

[0052] Step S101. Obtain real-time traffic data and customs clearance time prediction data, the real-time traffic data is obtained by calling a map application program interface, and is used to calculate congestion delay caused by road section congestion; and the customs clearance time prediction data is obtained by a customs declaration and inspection time prediction model, and is used to predict clearance delay caused by pre-audit of a declaration file, port congestion and inspection proportion.

[0053] Specifically, dynamic traffic information (such as real-time vehicle speed, congestion road section position and congestion level, traffic accident / construction event, etc.) is obtained through a map API, and is used to quantify changes in road section traffic efficiency.

[0054] The customs clearance time prediction data is obtained by constructing a customs declaration and inspection time prediction model, integrating parameters such as pre-audit state of a declaration file (such as file integrity, compliance), real-time flow of a port (historical same period data + real-time queuing vehicle number), and inspection probability (risk score based on cargo type and historical inspection record), and predicting delay of a clearance link.

[0055] The real-time traffic data is obtained by calling a third-party map API, pulling real-time speed and congestion index (such as a road section with speed lower than 40 km / h marked as congestion) of each road section on a transportation route at a preset frequency (such as every 5 minutes), and analyzing length and expected traffic speed of a congestion road section. A sudden traffic event (such as a traffic accident or temporary road closure) is structured and processed to generate an event influence range and expected recovery time.

[0056] The customs clearance prediction model construction includes: data input: declaration information (commodity type, declaration time, port code), historical port clearance data (average clearance time at different times, inspection rate), current port real-time queuing vehicle number (obtained through a port Internet of Things device or customs open data), and file state (such as “to be corrected” and “passed”) fed back by a pre-audit system. Gradient boosting tree (XGBoost) or random forest algorithm is used, the training target is to predict total clearance time (from arrival at a port to completion of inspection and release), and the features include file pre-audit state (coded as 0 / 1 / 2, representing different delay levels), port congestion index (mapped to 1-5 levels based on queuing vehicle number), and inspection probability (cargo type inspection rate calculated through historical data). The output result is a clearance delay prediction value (such as “estimated clearance time 90 minutes, including additional 45 minutes delay caused by 30% inspection possibility”).

[0057] Step S102. Based on the obtained real-time traffic data and customs clearance time prediction data, combined with line data of a map or a transportation system and average driving speed of a vehicle under normal conditions, a basic transportation time is calculated, and congestion delay and clearance delay are added to generate an initial path transportation time.

[0058] The base transportation time is calculated based on the route distance (the shortest path or specified route distance obtained through a map API from the starting point to the ending point) and the average driving speed of the vehicle under normal conditions (e.g., 80 km / h on a highway, 60 km / h on a regular road, adjusted according to the vehicle type). The congestion delay is calculated based on the difference between the actual speed of the congested section in real-time traffic data and the normal speed, and the additional time consumed (e.g., a certain section normally takes 10 minutes, but in real-time it takes 25 minutes, so the congestion delay is 15 minutes). The customs clearance delay directly uses the customs clearance time predicted by the model in S101.

[0059] The route segmentation processing segments the transportation route according to road types (highway / regular road / port access), national / territorial boundaries (to distinguish between domestic and foreign sections), and calls the map API to obtain the distance and normal speed limit of each segment. For example, a domestic highway section is 200 km long with a normal speed of 80 km / h, and the base time is 2.5 hours; a foreign regular road section is 150 km long with a normal speed of 50 km / h, and the base time is 3 hours.

[0060] The delay calculation includes: congestion delay: for real-time congested sections, calculate according to the formula Congestion Delay = Section Distance / Real-Time Speed - Section Distance / Normal Speed, and accumulate the delay of each congested section. Customs clearance delay: directly use the predicted time output by the customs model (e.g., including the total time of pre-examination delay, inspection waiting, etc.).

[0061] Initial path time integration: Total Time = Sum of Base Times of Each Segment + Total Congestion Delay + Customs Clearance Delay. For example: Base Time 5.5 hours + Congestion Delay 0.5 hours + Customs Clearance Delay 1.5 hours = Initial Transportation Time 7.5 hours.

[0062] Step S103. Obtain the driver's historical driving data, including average speed, brake frequency, and sudden acceleration, build a personalized correction coefficient library based on the historical driving data, and correct the initial path transportation time according to the personalized correction coefficient library to obtain the corrected transportation time.

[0063] Specifically, driver historical data collection: Obtain the driver's historical driving data through the vehicle-mounted GPS device or TBox (remote information processing box), including average speed (the ratio of actual driving speed to road speed limit) and sudden acceleration / brake frequency (reflecting driving stability).

[0064] Correction coefficient library construction: Establish a correlation model between driver driving behavior and transportation time, and generate personalized correction coefficients (e.g., "drivers who frequently accelerate suddenly spend 10% more time on the same section than usual").

[0065] Data collection and preprocessing collects GPS trajectory, speed, acceleration data of each transport through OBD interface or on-board sensors, labels sudden acceleration (acceleration > 0.3g) and sudden braking (deceleration <-0.3g) events, and calculates the number of sudden acceleration / braking per 100 kilometers. The ratio of the driver's historical average speed to the regular speed of the line is calculated (for example, if a driver's average speed is 90% of the regular speed, it is marked as "slow").

[0066] The correction model construction includes a rule engine or a machine learning model: rule engine: set rules such as "sudden acceleration / braking frequency > 20 times per 100 kilometers, transport time increased by 5%; average speed is 10% lower than the regular speed, time increased by 8%". Regression model: label historical transport time deviation, input driver behavior characteristics (average speed ratio, sudden acceleration frequency, braking frequency), train linear regression model, output correction coefficient (for example, coefficient = 1.12 represents a 12% increase in time).

[0067] Time correction: corrected transport time = initial path transport time x personalized correction coefficient. Example: initial time 7.5 hours, a driver's correction coefficient 1.08 (due to slower average speed), corrected time 8.1 hours.

[0068] Step S104. Obtain multi-dimensional parameter data, including vehicle type, route, weather, holiday, port flow and driver score, input the multi-dimensional parameters into the intelligent time prediction model trained based on historical transport data, generate the initial ETA; real-time acquisition of node reporting data and traffic congestion data in the transport process, dynamic correction of the initial ETA, output the final corrected ETA.

[0069] Specifically, multi-dimensional parameter input includes: integrated vehicle type (affects load and speed limit, such as truck vs car), route characteristics (slope, bend density), weather (rain / snow causes speed limit), holiday (port closure or congestion), port flow (real-time queue number), driver score (historical punctuality rate) and other parameters. The intelligent time prediction model is trained based on historical transport data to output the initial ETA; during transport, the ETA is dynamically adjusted through real-time node data (such as GPS position reporting, traffic events).

[0070] Multi-dimensional parameter processing includes: structured features: vehicle type (coded as categorical features), route characteristics (extract road segments with slope > 5° from GIS data), weather (call weather API to get real-time precipitation / wind speed, convert to speed limit coefficient, such as 20% speed limit reduction on rainy days), holiday (label whether it is a statutory holiday in the country where the port is located, affecting customs clearance time), driver score (0-5 points, reflecting historical time reliability). Data normalization: standardize continuous features (such as port flow), and one-hot encode categorical features (such as vehicle type).

[0071] Initial ETA generation includes model training: using historical transportation data (including actual arrival time, multi-dimensional parameters, final correction time) to train LSTM, LightGBM or deep learning model, input multi-dimensional parameters and corrected transportation time, output initial ETA (estimated arrival timestamp). Example: input vehicle type (heavy truck), route (including 20% mountainous road section), weather (light rain, speed limit 10%), driver rating 4.5 points, model output initial ETA is "2025-08-29 15:00".

[0072] Dynamic correction mechanism: real-time node data: report location every 15 minutes through vehicle-mounted GPS, calculate the deviation of current traveled distance and planned distance (such as "traveled 100 km, should travel 120 km, lag 20 km"); receive traffic event push (such as temporary closure of the front road section, need to detour 10 km). Correction algorithm: use Kalman filter or dynamic time series model to update remaining distance prediction combined with real-time deviation, formula is: remaining time = remaining distance / real-time average speed × dynamic correction factor (considering congestion changes) Output final ETA: after each node report, recalculate the remaining time based on the current location and real-time data, update the initial ETA and output.

[0073] In some embodiments, the obtained real-time traffic data and customs clearance time prediction data are combined with map or transportation system route data and average driving speed of the vehicle under normal conditions to calculate the basic transportation time, including: obtaining the total distance of the transportation route from the map or transportation system, determining the average driving speed of the vehicle under normal conditions based on historical operation data, vehicle type characteristics or industry experience value, dividing the total distance by the average driving speed to obtain the basic driving time without considering traffic and customs clearance factors; the real-time traffic data is used to calculate the increased driving time of each road section due to congestion as congestion delay, and the customs clearance time prediction data is used to determine the customs clearance delay corresponding to the pre-audit state of the customs declaration file, the degree of port congestion and the inspection probability.

[0074] The embodiments clarify the calculation logic of the basic transportation time, including calculating the basic driving time based on the total distance of the route and the average driving speed under normal conditions, quantifying the congestion delay combined with real-time traffic data, and determining the customs clearance delay through customs clearance prediction data. The core is to split the transportation process into "uninterrupted basic time" and "external factor delay" two parts, and calculate them respectively and then superimpose them.

[0075] Basic driving time calculation includes: total distance of the route: through the map API or the built-in route planning module of the transportation system, obtain the total distance of the actual driving route from the starting point to the ending point (accurate to meters), support custom route (such as avoiding mountainous area, preferential port channel).

[0076] Conventional average driving speed determination includes: historical operation data: statistics of the same vehicle model and the same line in the past 30 days average driving speed (extreme weather / congestion influence is excluded), such as the conventional speed of a certain line is 65km / h. Vehicle characteristics: adjustment according to vehicle load / axle number, for example, the conventional speed of heavy truck is set to 60km / h, and the conventional speed of light truck is set to 75km / h. Industry experience value: reference to logistics industry standard, highway conventional speed 80km / h, ordinary road 60km / h, rural road 40km / h. Formula: basic driving time = total distance ÷ conventional average driving speed (unit unified as km / h and hour).

[0077] Congestion delay calculation includes: real-time traffic data analysis: get real-time speed, congestion level of each road section from map API, calculate single section delay for congestion road section (speed < conventional speed 80%) according to congestion delay = road section distance ÷ real-time speed - road section distance ÷ conventional speed, and accumulate all congestion road section delay.

[0078] Customs clearance delay determination through customs clearance prediction data includes customs declaration file pre-audit status (such as "to be corrected" increases 2 hours delay, "pass" no delay), port congestion degree (queuing vehicle number is mapped to delay coefficient, such as each increase 10 vehicles extend 15 minutes), inspection probability (historical inspection rate 30% corresponds to average additional 60 minutes inspection time), the three weighted sum gets customs clearance delay (such as pre-audit delay 30 minutes + congestion 45 minutes + inspection expected 18 minutes = total customs clearance delay 93 minutes).

[0079] In some embodiments, the driver historical driving data includes average speed, brake frequency, and sudden acceleration, and a personalized correction coefficient library is constructed based on the historical driving data, including: collecting the average speed, unit mileage brake frequency and sudden acceleration times of the driver in the historical transportation task through the vehicle-mounted sensor or the driver terminal device, for statistical analysis according to the preset time period, establishing the corresponding relationship between different driving behavior characteristics and transportation time correction coefficient, the correction coefficient is determined by the ratio of the actual arrival time to the basic transportation time in the historical transportation task, forming a personalized correction coefficient library for each driver.

[0080] Embodiments focus on driver historical driving data collection and personalized correction coefficient library construction, driver behavior characteristics (average speed, brake frequency, and sudden acceleration times) are collected through vehicle-mounted devices, statistical analysis is performed according to time period, a mapping relationship between driving behavior and transportation time correction coefficient is established, the correction coefficient is determined by the ratio of the actual time consumed to the basic time in the history, forming a correction coefficient library exclusive for the driver.

[0081] Data collection channels include: on-board sensors: real-time collection of vehicle speed, acceleration, GPS position through OBD interface or T-Box device, calculation of the number of rapid acceleration (acceleration > 0.3g) and rapid braking (deceleration <-0.3g) per kilometer, and statistics of the average speed per trip (excluding parking time). Driver terminal device: the driver manually reports special events (such as temporary rest, road conditions) through the APP to assist in data calibration.

[0082] Statistical analysis and coefficient calculation includes: time period: according to natural month or every 50 transportation tasks as the period, statistics of the average speed deviation of the driver on different lines (actual speed / regular speed), brake frequency per unit distance (times / 100km), and rapid acceleration frequency (times / 100km). Correction coefficient calculation: for each historical task, calculate the correction coefficient = actual arrival time ÷ basic driving time, take the average coefficient of the past N tasks of this driver (N≥20), for example, the historical average coefficient of a driver is 1.05, which means that his driving habit causes the time consumption to be 5% more than the regular time.

[0083] The correction coefficient library is stored with driver ID as the index, and stores the correction coefficients corresponding to each feature, such as {driver ID: {average speed ratio: 0.92, brake frequency: 1.03, rapid acceleration frequency: 1.04, comprehensive correction coefficient: 1.05}}, and supports real-time updating (automatically synchronize the coefficient after the completion of a new task).

[0084] In some embodiments, the initial path transportation time is corrected according to the individualized correction coefficient library to obtain a corrected transportation time, including: according to the identity of the driver currently performing the transportation task, matching the corresponding driving behavior correction coefficient from the individualized correction coefficient library, multiplying the initial path transportation time by the correction coefficient, if it is detected that there is a risk driving behavior such as rapid acceleration frequency exceeding a preset threshold, continuous rapid braking, etc., increasing a preset risk delay based on the correction result, forming a corrected transportation time considering individual differences of the driver.

[0085] The embodiments describe how to correct the initial time using the correction coefficient library, including matching the driver's exclusive correction coefficient, and additionally increasing a preset delay for risk driving behavior (rapid acceleration exceeding the threshold, continuous rapid braking), to realize fine adjustment of individual differences of the driver.

[0086] The correction coefficient matching confirms the current driver identity through the driver login account, the on-board device binding ID, or face recognition, and retrieves the corresponding comprehensive correction coefficient (such as the comprehensive coefficient of driver A is 1.08) from the correction coefficient library. Formula: corrected basic time = initial path transportation time × comprehensive correction coefficient (excluding congestion / toll delay).

[0087] Risk driving behavior detection and delay overlay: threshold setting: preset threshold of frequent rapid acceleration (e.g. 20 times / 100km), continuous rapid braking judgment condition (3 times / 5 minutes). Real-time detection: real-time monitoring of driving behavior through vehicle-mounted sensors, if the frequency of rapid acceleration in the current transportation exceeds the threshold, increase 5 minutes of delay for every 5 times of exceeding the threshold; when continuous rapid braking is triggered, increase 3 minutes of delay for each event (based on the average additional time caused by such behavior in historical data). Final revised time: base time after revision + congestion delay + customs delay + risk delay.

[0088] In some embodiments, the inputting the multi-dimensional parameters into the intelligent time efficiency prediction model trained based on historical transportation data to generate an initial ETA comprises: using historical transportation data as a training set, using vehicle type, route, weather, holiday, port flow and driver score as input features, and using actual arrival time as output label, and using a machine learning algorithm to train an intelligent time efficiency prediction model; the machine learning algorithm includes but is not limited to random forest, gradient boosting tree or neural network algorithm; inputting the multi-dimensional parameters of the current transportation task into the model to calculate and output the initial ETA considering multi-dimensional factors through the intelligent time efficiency prediction model.

[0089] The embodiments describe the training and application of the intelligent time efficiency prediction model, which uses historical transportation data (input features are multi-dimensional parameters, and output labels are actual arrival times), trains the model through a machine learning algorithm (random forest, gradient boosting tree, neural network, etc.), and inputs current task parameters to generate an initial ETA considering multi-dimensional factors.

[0090] Training data preparation: input features: vehicle type (categorical variable, one-hot encoding), route features (percentage of road sections with slope > 5°, curve density), weather (temperature, probability of precipitation, converted to speed limit coefficient), holiday (binary variable, 1 = holiday), port flow (real-time queue number normalization), driver score (0-5 points, continuous variable). Output label: actual arrival timestamp (accurate to minutes), calculate time difference from planned departure time as regression target.

[0091] Model training process includes: algorithm selection: random forest: suitable for processing nonlinear relationships and categorical features, select split features through Gini coefficient, set tree depth 10-20 to solve overfitting. Gradient boosting tree (such as XGBoost): use mean square error (MSE) as loss function, set learning rate 0.1, iteration number 100, support feature importance analysis. Neural network: build fully connected layers, number of input layer neurons = number of features, 2-3 hidden layers, output layer for time difference prediction, use ReLU as activation function. Training and verification: use 5-fold cross-validation, evaluation index is mean absolute error (MAE), require MAE≤30 minutes.

[0092] Initial ETA generation includes inputting current task multi-dimensional parameters, model outputting predicted time difference, and combining departure timestamp to calculate initial ETA (such as departure time 2025-08-29 08:00, model predicting time consumption of 8.5 hours, and initial ETA being 16:30).

[0093] In some embodiments, the initial ETA is dynamically corrected by real-time acquisition of node reported data and traffic congestion data during transportation, and a final corrected ETA is output, including: real-time reporting of the current position of the vehicle and the time of arriving at the preset key node by the vehicle-mounted GPS device or the driver terminal, acquiring real-time updated traffic congestion data and port customs status change data, and recalculating the congestion delay of the current road section and the customs delay of the remaining road section based on the newly acquired traffic congestion data and port customs status change data; combining the remaining transportation distance and the real-time driving behavior data of the driver to iteratively update the initial ETA and output the final corrected ETA containing real-time state.

[0094] The embodiment defines a dynamic correction mechanism, recalculates the congestion / customs delay of the remaining road section through real-time node data (GPS position, key node arrival time), the latest traffic / customs status, iteratively updates the initial ETA by combining the remaining distance and real-time driving behavior, and outputs the final corrected ETA containing real-time state.

[0095] Real-time data acquisition includes: node reporting: preset key nodes (such as port entrance and highway service area) in the transportation route, vehicle-mounted GPS reports position every 15 minutes, judges whether to arrive at the node and records the time (such as planned to arrive at A node at 10:00, actually arrives at 10:15, lags behind for 15 minutes). Data acquisition: real-time pulling of traffic API to acquire the latest congestion status of the remaining road section, and calling the customs system to acquire the real-time queue number and inspection progress of the port (such as the current customs delay being updated from 90 minutes to 120 minutes).

[0096] Dynamic correction calculation includes: remaining distance calculation: total distance - distance traveled, combining real-time average speed (average speed of the traveled road section) to estimate the basic time of the remaining road section. Delay update: recalculating the congestion delay of the remaining road section according to new congestion data, and adjusting the customs delay according to the latest status of the port (such as the inspection probability being increased from 30% to 40%, increasing the delay by 20 minutes). Driving behavior correction: real-time monitoring whether the current average speed of the driver deviates from the historical average, and dynamically adjusting the correction coefficient (such as the current speed being 10% faster than the regular speed, and the correction coefficient being temporarily adjusted to 1.02).

[0097] The iterative updating algorithm adopts a sliding window model. After each node report, the remaining ETA is calculated as current time + remaining distance ÷ real-time speed + latest congestion delay + latest customs clearance delay × real-time correction factor, and the updated ETA is output (e.g., refreshed every 30 minutes).

[0098] In some embodiments, the method further comprises: when the time difference between the initial ETA output by the intelligent time limit prediction model and the promised time limit exceeds a preset threshold, triggering a multi-end collaborative early warning mechanism, and pushing a notification containing the real-time predicted time limit, delay risk factors and alternative solutions to the customer side; if the number of consecutive rapid accelerations or the frequency of rapid braking in the driver's driving behavior data exceeds the risk threshold, the system automatically sends a driving behavior warning to the driver's end and synchronously adjusts the risk delay in the corrected transportation time, forming a dynamic time limit correction mechanism containing risk intervention.

[0099] The embodiments add two mechanisms: ① Time limit warning mechanism: when the time difference between the initial ETA and the promised time limit exceeds the threshold, trigger multi-end collaborative warning, and push delay information and solutions to the customer; ② Risk intervention mechanism: when the driver's risky driving behavior is detected, send a driving warning and adjust the risk delay, forming a closed-loop correction.

[0100] The time limit warning mechanism includes: threshold setting: according to the logistics contract, set the warning threshold (e.g. promised time limit ± 45 minutes, exceeding triggers warning). Multi-end collaboration: customer side: push warning information through SMS and APP, including current ETA, delay reason (e.g. "port congestion causes 60 minutes delay"), alternative solution (e.g. suggest partial shipment, adjust delivery time). Dispatch side: generate dispatch work order, prompt to arrange backup vehicles or coordinate port priority customs clearance. Data synchronization: after the warning is triggered, the system automatically records the delay reason and synchronizes it to the historical data for model optimization. The risk intervention mechanism includes: real-time detection: monitor the frequency of rapid acceleration / braking through the vehicle-mounted sensor in real time, and determine risky driving when rapid braking ≥ 3 times or rapid acceleration ≥ 4 times within 5 minutes. Intervention measures: driver's end: send warning through vehicle-mounted screen or voice broadcast (e.g. "please drive smoothly, current driving behavior may cause delay"). Time limit correction: increase the preset risk delay (e.g. increase 2 minutes for each excessive rapid acceleration, cumulative not more than 30 minutes) immediately when each risk event triggers, and recalculate the ETA. Closed-loop feedback: if the driver's behavior improves after receiving the warning, the system automatically reduces the subsequent risk delay, forming a "detection-warning-correction" dynamic cycle.

[0101] In some embodiments, for complex scenarios of cross-border multimodal transportation (road + sea / rail), a graph neural network model is constructed, the transportation network is abstracted into a graph structure of nodes (cities, ports, transit stations) and edges (road sections, shipping routes, customs clearance processes), and the dependence between different transportation modes and port nodes (such as the conduction effect of port congestion on road connection time) is captured, solving the problem that traditional models are difficult to handle multi-modal coupling effects.

[0102] The graph structure definition includes: node features: city nodes (population density, transportation hub level), port nodes (inspection efficiency, berth quantity), transit station nodes (warehouse capacity, sorting efficiency); edge features (real-time congestion coefficient of road sections, weather influence factor of sea transportation routes, punctuality rate of railway schedules).

[0103] The GNN model architecture includes: graph convolution layer (GCN): aggregate adjacent node features (such as calculating the customs clearance delay of a port node, considering the congestion state of its upstream road node and the berth occupancy rate of the downstream port). Attention mechanism: design "modal attention head" to dynamically weight the interaction effects of different transportation modes such as road, sea, etc. (such as reducing the weight of road transportation and increasing the reliability assessment of railway transportation under heavy rain).

[0104] The training and inference process includes: data input: convert multimodal transportation orders into graph structures, input time efficiency data of each node in historical transportation (such as actual driving time of road sections, port loading and unloading time), and label the final delivery time deviation. Prediction output: for cross-border orders containing road-sea connection, output the predicted arrival time of each key node (such as "expected to arrive at the port transit station at 14:00, and complete loading at 16:30"), and back-propagate the overall ETA.

[0105] Support "broken link prediction": when a section of road is suddenly closed, the model automatically calculates the time efficiency compensation by switching transportation modes through adjacent transit stations (such as from road to railway, predicting that the transit station sorting delay increases by 40 minutes, and the overall ETA delays by 1.5 hours).

[0106] In some embodiments, for the privacy protection needs of driver driving data, a federated learning framework is used to aggregate the driver behavior data of multiple logistics companies to train personalized models without sharing original data. Each enterprise locally trains a driver behavior correction coefficient model, and improves the model generalization ability through encrypted parameter interaction, solving the problems of insufficient data volume and privacy compliance of a single enterprise.

[0107] The federated learning architecture includes: participants: logistics enterprises A, B, and C (each has its own driver driving data, including the number of sudden accelerations and brake frequencies, but does not share the original trajectory). Central server: coordinates model parameter aggregation and maintains a global driver behavior feature library (such as the average correction coefficient distribution of "high-frequency sudden acceleration drivers").

[0108] Local model training uses self-owned data to train LSTM models locally by each enterprise, inputs the driver's historical behavior sequence (sudden acceleration / brake frequency sequence divided by time window), and outputs the driver's real-time correction coefficient prediction (such as the time consumption increase caused by driving style changes in the next 2 hours).

[0109] After the local model parameters are encrypted and uploaded (such as using homomorphic encryption technology), the central server uses differential privacy protection when aggregating to avoid revealing single-enterprise data features. When new drivers are hired, if the local data of the enterprise is insufficient, the "industry-level driving behavior prior distribution" is obtained through the federated model, and the correction coefficient is quickly calibrated combined with the driver's first 3 transportation data (such as the initial coefficient using the industry average of 1.07, and then fine-tuned to 1.05 through local data). Data interaction only transmits model gradients or aggregated feature statistics (such as mean and variance), and the original driving trajectory data is always stored locally in the enterprise, which complies with data protection regulations such as GDPR.

[0110] In some embodiments, for low-probability high-impact scenarios such as typhoons, snowstorms, and sudden public events such as port strikes, a large amount of virtual extreme scenario data is generated using a generative adversarial network to enhance the generalization ability of the intelligent time-sensitive prediction model. Through the "generator" to simulate the traffic and customs abnormal state under rare scenarios, the "discriminator" to distinguish real and virtual data, and to improve the prediction robustness of the model in unknown scenarios.

[0111] The data generation process includes: Generator (G): input random noise vector, combined with the characteristics of historical extreme events (such as typhoon path, strike duration), output traffic data under virtual scenarios (such as a road segment speed limit 20km / h in a typhoon, port closed for 12 hours). Discriminator (D): receives real extreme scenario data (such as customs records under the influence of historical typhoons) and generated data, and improves the authenticity of generated data through adversarial training (such as the K-S test value of the generated port closure duration distribution and the real event is <0.1).

[0112] The enhanced training strategy includes: mixed training set: mix the generated extreme scenario data with normal scenario data at a ratio of 1:9, input the intelligent time prediction model (such as improved LightGBM), and increase the "extreme scenario penalty term" (such as the prediction error weight in the loss function is increased by 3 times) in the loss function. Scene annotation: add scene labels (such as "typhoon influence" and "port strike") to each generated data, and the model learns the time impact pattern of different abnormal types (such as strikes mainly affect the customs pre-audit link, and typhoons mainly affect road driving speed) during training.

[0113] Real-time inference application includes: when the meteorological department issues a typhoon warning, trigger the extreme scenario inference module, generate virtual speed limit data and port temporary closure plan for each road segment under the typhoon path, input the model and output "extreme scenario ETA" (such as normal ETA is 16:00, and after inference, it is adjusted to 19:30, and marked "typhoon influence: road speed limit causes 2.5 hours delay, port temporary closure 1 hour").

[0114] In some embodiments, for the difference in influence degree of multi-dimensional parameters (traffic, customs, driver, weather) under different scenarios, a self-attention mechanism is introduced into the intelligent time prediction model, so that the model can dynamically allocate the weight of each parameter (such as automatically increasing the weight of "weather" parameter under heavy rain, and reducing the weight of "driver score"), solving the problem that the fixed weight of traditional model cannot adapt to scenario changes.

[0115] Attention layer design includes: query (Query): key features of the current transportation scenario (such as port type, cargo temperature requirement). Key (Key): feature vector of each parameter (such as traffic congestion index is encoded as [0.8, 0.3, 0.1], indicating the congestion state of highway / ordinary highway / port channel). Value (Value): quantitative value of the influence of each parameter on time (such as the delay minutes caused by traffic congestion, the probability value of customs inspection).

[0116] The weight calculation logic calculates the similarity between Query and Key through dot product, and generates attention weights through Softmax (for example, in a heavy rain scenario, the weight of the "weather" parameter increases from 0.15 to 0.35, and the weight of the "driver historical speed" parameter decreases from 0.2 to 0.15). The weighted aggregation of the Value values of each parameter obtains the dynamically corrected time-sensitive influence factor (for example, the formula: total delay = Σ(weight_i x delay_contribution_i)). Model integration and training embed attention mechanisms into LightGBM or Transformer models, and add a "scene classification auxiliary task" (such as distinguishing between daily, holiday, and extreme weather scenes) during training to guide the optimization of attention weights in scene-sensitive directions. During real-time inference, the attention head of the corresponding parameter is automatically activated according to the current scene (such as automatically enhancing the attention of the "port traffic" parameter during the cross-border e-commerce promotion season). In complex scenarios (such as holiday overlapping with heavy rain), the parameter weight distribution accuracy is improved, and the ETA prediction error is reduced compared to the fixed weight model.

[0117] In some embodiments, a digital twin system is constructed for a cross-border transportation network, which dynamically simulates the impact of different decisions on time efficiency in virtual space by real-time mapping of vehicle positions, traffic conditions, and port operation processes in the physical world (such as the time efficiency gain of informing the driver to switch routes 30 minutes in advance). Combined with the simulation results of the digital twin, real-time correction strategies are optimized to realize a "prediction-simulation-verification" closed loop.

[0118] The twin model construction includes: physical layer: collect vehicle GPS, port Internet of Things devices (such as scale, X-ray machine status), and real-time traffic camera images. Virtual layer: use Unity / UE engine to build 3D transportation network, and synchronize physical status of each road segment and port virtual entity (such as the virtual queuing vehicle number of the port inspection channel is consistent with the actual number).

[0119] The real-time simulation process synchronizes physical data to the twin system every 10 minutes, and the virtual vehicle travels according to the current driver's driving behavior (such as the frequency of sudden acceleration) and real-time traffic conditions, simulating the time of arrival at each node. When the path needs to be adjusted (such as congestion ahead), the time efficiency results of 2-3 alternative routes are simulated in parallel in the twin system (such as route 1 is expected to arrive at 15:30, and route 2 is expected to arrive at 15:15 but increases the distance by 50km), and the optimal solution is output to the actual system.

[0120] The closed-loop optimization mechanism includes: comparing the actual transportation results with the twin simulation results, updating the parameters of the virtual model (such as the actual inspection efficiency of a port is 10% faster than the historical data, correcting the inspection time parameter of the twin). Use the twin system to conduct stress testing (such as simulate the impact of the sudden closure of a port for 4 hours), develop emergency plans in advance and update the abnormal handling rules of the time prediction model. Realize the "what you see is what you get" time deduction, support the dispatcher to pre-verify the decision in the virtual space, reduce the trial and error cost of actual adjustment, and improve the decision response speed in extreme scenarios.

[0121] In some embodiments, in the cross-border whole vehicle (FTL) transportation business, the uncertainty of time has always been a pain point. The current main problems are: inaccurate path planning: traditional systems are only based on distance or static rules, and cannot integrate real-time traffic and port customs clearance, resulting in large deviation of ETA. Driver individual differences are not modeled: existing solutions ignore the impact of driver driving habits on time, and fail to achieve personalized correction. Lack of multi-end collaboration: customers can only see the promised time, and drivers only have regular tasks, with no two-way linkage, lack of exception visualization and compensation mechanism. The prediction model is rough: the existing ETA prediction accuracy is low, and cannot combine weather, holidays, cargo type and port flow with multiple factors, with an error often exceeding 1 hour.

[0122] Cross-border whole vehicle business involves trunk transportation, customs declaration, port customs clearance, and delivery of multiple links. Existing transportation management systems are mostly based on experience rules and fixed tables, lacking dynamic adjustment capability. A small number of systems that use external map APIs also only consider distance and road conditions, and fail to integrate customs models and holiday delays. ETA prediction still relies on mean value algorithm, and cannot reflect driver differences and route characteristics. Therefore, a new intelligent system that integrates real-time traffic, customs rules, driver behavior and multi-end visualization is needed to improve the stability and customer experience of cross-border whole vehicle business.

[0123] The embodiment proposes a "capacity building system based on cross-border vehicle service", which can: use a hybrid path planning algorithm, integrate real-time traffic and customs models, and dynamically calculate ETA; Through the driver behavior learning model, realize personalized ETA correction and risk driving warning; Support multi-terminal collaborative visualization, customer side double time axis, driver side key node reminder; Build an intelligent time limit prediction engine, support multi-dimensional parameter input, realize ETA error ≤ ± 15 minutes. Real-time traffic data is obtained by calling a map API (such as Tencent Map). Introduce the customs clearance time model to predict the delay of declaration / inspection. Comprehensive calculation of ETA: ETA=D / Vavg+Ttraffic+Tcuroms, wherein ETA represents the total distance of the transportation path (kilometers), which is calculated by the line data of the map API or the transportation system; D / Vavg represents the average driving speed of the vehicle under normal conditions (km / h), which can be set based on historical operation data, vehicle characteristics or industry experience value; represents the theoretical basic driving time (hours), that is, the ideal transportation time without considering traffic and customs factors; Ttraffic represents the delay caused by real-time traffic conditions (hours), which is obtained by real-time traffic flow through a map API (such as Tencent Map / Gaode Map) to estimate the increased driving time of the road section; Ttraffic represents the delay caused by real-time traffic conditions (hours), which is obtained by real-time traffic flow through a map API (such as Tencent Map / Gaode Map) to estimate the increased driving time of the road section; Tcuroms represents the delay that may be caused by customs clearance (hours), which is obtained by a customs declaration and inspection time prediction model, considering port congestion, customs document pre-audit situation and inspection proportion. The path can be dynamically adjusted according to different types of goods (general goods / hazardous goods) and policy rules.

[0124] Collect driver's historical driving data (average speed, brake frequency, sudden acceleration). Build a personalized correction coefficient library: ETA = baseline time × driving coefficient. High-risk driving behavior triggers an early warning and automatically extends ETA.

[0125] Client: Show "commitment time limit" and "real-time predicted time limit" double time axis, if overtime, push compensation scheme. Driver side: Embedded countdown timer, push preparation reminder at key nodes (such as 50km from the port). Management side: Unified visualization board, aggregate ETA deviation, risk behavior and delay reason.

[0126] Train the model based on historical transportation data, input features include: vehicle type, route, weather, holiday, port traffic, driver rating. Generate initial ETA, control error rate within ± 15 minutes. Dynamically correct ETA with real-time data (node reporting, traffic congestion).

[0127] Please refer to Figure 2As shown, Figure 2 is a structural schematic diagram of an ETA calculation device 200 for cross-border whole vehicle business provided by an embodiment of the present application. The ETA calculation device 200 for cross-border whole vehicle business is used to execute the steps of the ETA calculation method for cross-border whole vehicle business shown in each of the above embodiments. The ETA calculation device 200 for cross-border whole vehicle business can be a single server or a server cluster, or the ETA calculation device 200 for cross-border whole vehicle business can be a terminal, which can be a handheld terminal, a notebook computer, a wearable device, or a robot, etc.

[0128] As shown, Figure 2 The ETA calculation device 200 for cross-border whole vehicle business includes:

[0129] A data acquisition unit 201 is configured to acquire real-time traffic data and customs clearance time prediction data. The real-time traffic data is acquired by calling a map application interface, and is used to calculate congestion delay caused by road congestion. The customs clearance time prediction data is obtained by a customs declaration and inspection time prediction model, and is used to predict clearance delay caused by customs declaration pre-audit, port congestion and inspection ratio.

[0130] A time generation unit 202 is configured to calculate basic transportation time based on the acquired real-time traffic data and customs clearance time prediction data, in combination with line data of a map or a transportation system and average driving speed of a vehicle under normal conditions, and to generate initial path transportation time by adding congestion delay and clearance delay.

[0131] A time correction unit 203 is configured to acquire driver historical driving data including average speed, brake frequency and sudden acceleration condition, to construct a personalized correction coefficient library based on the historical driving data, and to correct the initial path transportation time according to the personalized correction coefficient library to obtain corrected transportation time.

[0132] A dynamic correction unit 204 is configured to acquire multi-dimensional parameter data including vehicle model, line, weather, holiday, port flow and driver score, to input the multi-dimensional parameter data into an intelligent time limit prediction model trained based on historical transportation data to generate an initial ETA, to acquire node reporting data and traffic congestion data in real time during transportation, and to dynamically correct the initial ETA to output a final corrected ETA.

[0133] In some embodiments, the base transportation time is calculated based on the acquired real-time traffic data and customs clearance time prediction data, combined with map or transportation system line data and average driving speed of the vehicle under normal conditions, including: obtaining the total distance of the transportation line from the map or transportation system, determining the average driving speed of the vehicle under normal conditions based on historical operation data, vehicle model characteristics or industry experience value, and dividing the total distance by the average driving speed to obtain the base driving time without considering traffic and customs clearance factors; the real-time traffic data is used to calculate the increased driving time of each road section due to congestion as congestion delay, and the customs clearance time prediction data is used to determine the customs clearance delay corresponding to the pre-audit status of the customs declaration file, the degree of port congestion and the inspection probability.

[0134] In some embodiments, the driver historical driving data includes average speed, brake frequency and sudden acceleration, and the personalized correction coefficient library is constructed based on the historical driving data, including: collecting the average speed, brake frequency per mile and sudden acceleration times of the driver in the historical transportation task through the vehicle-mounted sensor or the driver terminal device, for statistical analysis according to the preset time period, establishing the corresponding relationship between different driving behavior characteristics and transportation time correction coefficients, the correction coefficient is determined by the ratio of the actual arrival time to the base transportation time, and the personalized correction coefficient library for each driver is formed.

[0135] In some embodiments, the initial path transportation time is corrected according to the personalized correction coefficient library to obtain the corrected transportation time, including: according to the identity of the driver currently performing the transportation task, matching the corresponding driving behavior correction coefficient from the personalized correction coefficient library, multiplying the initial path transportation time by the correction coefficient, if it is detected that the sudden acceleration frequency exceeds the preset threshold, continuous sudden braking and other risk driving behaviors exist in the current driving behavior, increasing the preset risk delay based on the correction result, forming the corrected transportation time considering the individual differences of the driver.

[0136] In some embodiments, the multi-dimensional parameters are input into the intelligent time prediction model trained based on historical transportation data to generate the initial ETA, including: using historical transportation data as a training set, using vehicle model, line, weather, holiday, port flow and driver score as input features, and using actual arrival time as output label, and using machine learning algorithm to train to obtain the intelligent time prediction model; the machine learning algorithm includes but is not limited to random forest, gradient boosting tree or neural network algorithm; the multi-dimensional parameters of the current transportation task are input into the model, and the initial ETA considering multi-dimensional factors is calculated and output by the intelligent time prediction model.

[0137] In some embodiments, the real-time acquisition of node reporting data and traffic congestion data during transportation, dynamic correction of the initial ETA, and output of the final corrected ETA include: real-time reporting of the current position of the vehicle and the time of arrival at the preset key node through the vehicle-mounted GPS device or the driver terminal, obtaining real-time updated traffic congestion data and port customs status change data, and recalculating the congestion delay of the current road section and the customs delay of the remaining road section based on the newly obtained traffic congestion data and port customs status change data; combining the remaining transportation distance and the real-time driving behavior data of the driver, iteratively updating the initial ETA, and outputting the final corrected ETA containing the real-time state.

[0138] In some embodiments, the method further includes: when the time difference between the initial ETA output by the intelligent time limit prediction model and the promised time limit exceeds a preset threshold, triggering a multi-end collaborative early warning mechanism, and pushing a notification containing the real-time predicted time limit, the delay risk factors and the alternative solutions to the customer side; if the number of consecutive rapid accelerations or the frequency of rapid braking in the driver's driving behavior data exceeds the risk threshold, the system automatically sends a driving behavior warning to the driver side and synchronously adjusts the risk delay in the corrected transportation time, forming a dynamic time limit correction mechanism containing risk intervention.

[0139] It should be noted that, for the convenience and brevity of description, the specific working process of the ETA calculation device and each module of the cross-border whole vehicle business described above can refer to the corresponding process in the ETA calculation method embodiments of the cross-border whole vehicle business described above, which will not be described here.

[0140] The above-mentioned ETA calculation method of the cross-border whole vehicle business can be implemented in the form of a computer program, which can run on the device as shown in Figure 2 .

[0141] Please refer to Figure 3 , Figure 3 is a structural schematic block diagram of a computer device provided by the embodiments of the present application. The computer device includes a processor, a memory and a network interface connected through a device bus, wherein the memory can include a storage medium and an internal memory.

[0142] The storage medium can store the operating device and the computer program. The computer program includes program instructions, which, when executed, can cause the processor to execute any kind of ETA calculation method of the cross-border whole vehicle business.

[0143] The processor is used to provide computing and control capabilities to support the operation of the entire computer device.

[0144] The internal memory provides an environment for the running of a computer program in a non-volatile storage medium, which, when executed by the processor, can enable the processor to perform any ETA calculation method for cross-border whole vehicle business.

[0145] The network interface is used for network communication, such as sending assigned tasks. Figure 3 The structure shown in the figure is only a block diagram of part of the structure related to the scheme of the present application, and does not constitute a limitation on the terminal to which the scheme of the present application is applied. A specific computer device can include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0146] It should be understood that the processor can be a central processing unit (CPU), and the processor can also be other general-purpose processors, digital signal processors (DSP), application specific integrated circuits (ASIC), field programmable gate arrays (FPGA) or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components, etc. The general-purpose processor can be a microprocessor or the processor can also be any conventional processor.

[0147] In one embodiment, the processor is configured to run a computer program stored in the memory to perform the following steps:

[0148] Real-time traffic data is obtained by calling a map application program interface, and is used to calculate congestion delay caused by road congestion; and customs clearance time prediction data is obtained by a customs declaration and inspection time prediction model, and is used to predict clearance delay caused by customs declaration pre-examination, port congestion and inspection ratio;

[0149] Based on the obtained real-time traffic data and customs clearance time prediction data, combined with line data of a map or a transportation system and average driving speed of a vehicle under normal conditions, a basic transportation time is calculated, and congestion delay and clearance delay are added to generate an initial path transportation time;

[0150] Driver historical driving data including average speed, brake frequency and sudden acceleration are obtained, and a personalized correction coefficient library is constructed based on the historical driving data. The initial path transportation time is corrected according to the personalized correction coefficient library to obtain a corrected transportation time.

[0151] Obtain multi-dimensional parameter data, including vehicle type, line, weather, holiday, port flow and driver score, input the multi-dimensional parameters into an intelligent time prediction model trained based on historical transportation data to generate an initial ETA; obtain node reporting data and traffic congestion data in real time during transportation, dynamically correct the initial ETA, and output the final corrected ETA.

[0152] In some embodiments, the base transportation time is calculated based on the obtained real-time traffic data and customs clearance time prediction data, combined with map or transportation system line data and average driving speed of the vehicle under normal conditions, including: obtaining the total distance of the transportation line from the map or transportation system, determining the average driving speed of the vehicle under normal conditions based on historical operation data, vehicle type characteristics or industry experience value, dividing the total distance by the average driving speed to obtain the basic driving time without considering traffic and customs clearance factors; the real-time traffic data is used to calculate the increased driving time of each section due to congestion as congestion delay, and the customs clearance time prediction data is used to determine the customs clearance delay corresponding to the customs declaration file pre-audit state, port congestion degree and inspection probability.

[0153] In some embodiments, the driver historical driving data includes average speed, brake frequency and sudden acceleration, and the personalized correction coefficient library is constructed based on the historical driving data, including: collecting the average speed, brake frequency per mile and sudden acceleration times of the driver in historical transportation tasks through vehicle-mounted sensors or driver terminal devices, for statistical analysis according to a preset time period, establishing the corresponding relationship between different driving behavior characteristics and transportation time correction coefficients, the correction coefficient is determined by the ratio of the actual arrival time to the base transportation time, and the personalized correction coefficient library for each driver is formed.

[0154] In some embodiments, the initial path transportation time is corrected according to the personalized correction coefficient library to obtain the corrected transportation time, including: according to the identity of the driver currently performing the transportation task, matching the corresponding driving behavior correction coefficient from the personalized correction coefficient library, multiplying the initial path transportation time by the correction coefficient, if it is detected that the sudden acceleration frequency in the current driving behavior exceeds the preset threshold, there are risk driving behaviors such as continuous sudden braking, increasing the preset risk delay based on the correction result, forming the corrected transportation time considering the individual differences of the driver.

[0155] In some embodiments, the inputting the multi-dimensional parameters into an intelligent time prediction model trained based on historical transportation data to generate an initial ETA comprises: using historical transportation data as a training set, using vehicle models, routes, weather, holidays, port traffic, and driver ratings as input features, using actual arrival times as output labels, and using a machine learning algorithm to train an intelligent time prediction model; the machine learning algorithm includes but is not limited to random forest, gradient boosting tree or neural network algorithm; inputting the multi-dimensional parameters of the current transportation task into the model to calculate the initial ETA considering the multi-dimensional factors by the intelligent time prediction model.

[0156] In some embodiments, the method further comprises: real-time acquisition of node reporting data and traffic congestion data during transportation, dynamic correction of the initial ETA, and output of a final corrected ETA, comprising: real-time reporting of the current position of the vehicle and the time of arrival at the preset key node by the vehicle-mounted GPS device or the driver terminal, real-time updating of the traffic congestion data and the port customs status change data, re-computation of the congestion delay of the current road section and the customs delay of the remaining road section based on the newly acquired traffic congestion data and the port customs status change data; iterative updating of the initial ETA based on the remaining transportation distance and the real-time driving behavior data of the driver, and output of the final corrected ETA containing the real-time state.

[0157] In some embodiments, the method further comprises: when the time difference between the initial ETA output by the intelligent time prediction model and the promised time exceeds a preset threshold, triggering a multi-end collaborative early warning mechanism and pushing a notification containing the real-time predicted time, the delay risk factors and the alternative solutions to the customer side; if the number of consecutive rapid accelerations or the frequency of rapid braking in the driver's driving behavior data exceeds the risk threshold, the system automatically sends a driving behavior warning to the driver's end and synchronously adjusts the risk delay in the corrected transportation time, forming a dynamic time correction mechanism containing risk intervention.

[0158] In the embodiments of the present application, a computer readable storage medium is also provided, which stores a computer program including program instructions. The processor executes the program instructions to implement the steps of the ETA calculation method for cross-border whole vehicle business provided by the above-mentioned embodiments.

[0159] The computer readable storage medium can be an internal storage unit of the computer device, such as a hard disk or a memory of the computer device. The computer readable storage medium can also be an external storage device of the computer device, such as a plug-in hard disk, a smart media card (SMC), a secure digital (SD) card, a flash card, etc.

[0160] The above description is provided as an enabling teaching of the application and is not intended to limit its scope in any way. Any modification of the application in keeping with the spirit thereof and any further applications thereof within the technical field known to those skilled in the art are to be construed as falling within the purview thereof.

Claims

1. A method for calculating ETA (Electronic Toll Collection) in cross-border vehicle manufacturing transactions, characterized in that, include: Real-time traffic data and customs clearance time prediction data are obtained. The real-time traffic data is obtained by calling the map application interface and is used to calculate the congestion delay caused by road congestion. The customs clearance time prediction data is obtained by the customs declaration and inspection time prediction model and is used to predict the clearance delay caused by the pre-review of customs declaration documents, port congestion and inspection ratio. Based on the acquired real-time traffic data and customs clearance time prediction data, combined with the route data of the map or transportation system and the average driving speed of vehicles under normal conditions, the basic transportation time is calculated, and congestion delay and customs clearance delay are added to generate the initial route transportation time. Acquire historical driving data of drivers, including average speed, braking frequency, and rapid acceleration. Based on the historical driving data, construct a personalized correction coefficient library, including: collecting the driver's average speed, braking frequency per unit mileage, and number of rapid accelerations in historical transportation tasks through vehicle sensors or driver terminal devices, and performing statistical analysis according to a preset time period to establish the correspondence between different driving behavior characteristics and transportation time correction coefficients. The correction coefficient is determined by the ratio of the actual arrival time to the basic transportation time in historical transportation tasks, forming a personalized correction coefficient library for each driver. Correct the initial route transportation time according to the personalized correction coefficient library to obtain the corrected transportation time, including: matching the corresponding driving behavior correction coefficient from the personalized correction coefficient library according to the identity of the driver currently performing the transportation task, multiplying the initial route transportation time by the correction coefficient, and if it is detected that the current driving behavior has a rapid acceleration frequency exceeding a preset threshold or continuous rapid braking risk driving behavior, a preset risk delay is added to the correction result to form a corrected transportation time that takes into account individual differences of drivers. The system acquires multi-dimensional parameter data, including vehicle type, route, weather, holidays, port traffic flow, and driver rating. These parameters are input into an intelligent timeliness prediction model trained on historical transportation data to generate an initial ETA. During transportation, real-time node-reported data and traffic congestion data are acquired to dynamically correct the initial ETA, outputting a final corrected ETA. This process includes: real-time reporting of the vehicle's current location and arrival time at preset key nodes via onboard GPS devices or driver terminals; acquisition of real-time updated traffic congestion data and port clearance status changes; recalculation of congestion delay for the current road segment and clearance delay for the remaining road segment based on the newly acquired traffic congestion data and port clearance status changes; and iterative updates of the initial ETA, combining the remaining transportation distance and real-time driver behavior data, to output a final corrected ETA that includes real-time status.

2. The method according to claim 1, characterized in that, The basic transportation time is calculated based on the acquired real-time traffic data and customs clearance time prediction data, combined with route data from maps or transportation systems and the average vehicle speed under normal conditions, including: The total distance of the transportation route is obtained from a map or transportation system. The average driving speed of the vehicle under normal conditions is determined based on historical operation data, vehicle type characteristics, or industry experience values. The total distance is divided by the average driving speed to obtain the basic driving time without considering traffic and customs clearance factors. The real-time traffic data is used to calculate the increased driving time of each road segment due to congestion as congestion delay. The customs clearance time prediction data is used to determine the customs clearance delay corresponding to the pre-examination status of customs declaration documents, the degree of port congestion, and the probability of inspection.

3. The method according to claim 1, characterized in that, The step of inputting the multidimensional parameters into an intelligent timeliness prediction model trained based on historical transportation data to generate an initial ETA includes: Using historical transportation data as a training set, and taking vehicle type, route, weather, holidays, port traffic and driver rating as input features, and actual arrival time as output label, a machine learning algorithm is used to train an intelligent timeliness prediction model; the machine learning algorithm includes, but is not limited to, random forest, gradient boosting tree or neural network algorithm. The model inputs the multidimensional parameters of the current transportation task and calculates and outputs the initial ETA that takes into account the multidimensional factors through the intelligent timeliness prediction model.

4. The method according to claim 1, characterized in that, The method further includes: When the time difference between the initial ETA output by the intelligent timeliness prediction model and the promised timeliness exceeds a preset threshold, a multi-terminal collaborative early warning mechanism is triggered, and a notification containing real-time predicted timeliness, delay risk factors, and alternative solutions is pushed to the customer. If the number of consecutive rapid accelerations or the frequency of sudden braking in the driver's driving behavior data exceeds the risk threshold, the system will automatically send a driving behavior warning to the driver and simultaneously adjust the risk delay in the corrected transportation time, forming a dynamic timeliness correction mechanism that includes risk intervention.

5. An ETA calculation device for cross-border vehicle manufacturing transactions, characterized in that, include: The data acquisition unit is used to acquire real-time traffic data and customs clearance time prediction data. The real-time traffic data is acquired by calling the map application interface and is used to calculate the congestion delay caused by road congestion. The customs clearance time prediction data is obtained by the customs declaration and inspection time prediction model and is used to predict the clearance delay caused by the pre-review of customs declaration documents, port congestion, and inspection ratio. The time generation unit is used to calculate the basic transportation time based on the acquired real-time traffic data and customs clearance time prediction data, combined with the route data of the map or transportation system and the average driving speed of vehicles under normal conditions, and add congestion delay and customs clearance delay to generate the initial route transportation time. The time correction unit is used to acquire the driver's historical driving data, including average speed, braking frequency, and rapid acceleration. Based on the historical driving data, a personalized correction coefficient library is constructed. This includes: collecting the driver's average speed, braking frequency per unit mileage, and rapid acceleration times in historical transportation tasks through onboard sensors or driver terminal devices, performing statistical analysis according to a preset time period, establishing a correspondence between different driving behavior characteristics and transportation time correction coefficients, wherein the correction coefficient is determined by the ratio of the actual arrival time to the basic transportation time in historical transportation tasks, forming a personalized correction coefficient library for each driver; and correcting the initial route transportation time according to the personalized correction coefficient library to obtain the corrected transportation time, including: matching the corresponding driving behavior correction coefficient from the personalized correction coefficient library according to the driver's identity currently performing the transportation task, multiplying the initial route transportation time by the correction coefficient; if it is detected that the current driving behavior has a rapid acceleration frequency exceeding a preset threshold or continuous rapid braking risk driving behavior, a preset risk delay is added to the correction result to form a corrected transportation time that takes into account individual differences of the driver. A dynamic correction unit is used to acquire multi-dimensional parameter data, including vehicle type, route, weather, holidays, port traffic flow, and driver rating. These multi-dimensional parameters are input into an intelligent timeliness prediction model trained on historical transportation data to generate an initial ETA. During transportation, real-time node-reported data and traffic congestion data are acquired to dynamically correct the initial ETA, outputting a final corrected ETA. This includes: real-time reporting of the vehicle's current location and arrival time at preset key nodes via onboard GPS devices or driver terminals; acquisition of real-time updated traffic congestion data and port clearance status change data; recalculation of congestion delay for the current road segment and clearance delay for the remaining road segment based on the newly acquired traffic congestion data and port clearance status changes; and iterative updates of the initial ETA by combining the remaining transportation distance and real-time driver behavior data to output a final corrected ETA including real-time status.

6. A computer device, characterized in that, The computer device includes a memory and a processor; The memory is used to store computer programs; The processor is configured to execute the computer program and, in executing the computer program, implement the method as described in any one of claims 1 to 4.

7. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, causes the processor to implement the method as described in any one of claims 1 to 4.

Images