Intelligent recommendation method and system for logistics scheduling path based on large model

Abstract

The application provides a logistics scheduling path intelligent recommendation method and system based on a large model. The method comprises the following steps: obtaining a logistics scheduling task image; generating a corresponding dynamic line familiarity based on the logistics scheduling task image; generating a logistics scheduling candidate scheme based on the dynamic line familiarity and in combination with a predicted arrival arrangement and a loading and unloading connection arrangement corresponding to a vehicle available state; inputting an input sequence constructed based on the logistics scheduling candidate scheme into a large model to perform a performance deduction according to an automobile logistics operation chain, and generating a logistics scheduling path recommendation result. The application makes the logistics scheduling path recommendation more suitable for the actual operation scene of the automobile logistics multi-link, strong time limit and strong connection.

Description

Technical Field

[0001] This invention belongs to the field of intelligent recommendation of logistics scheduling routes, and particularly relates to a method and system for intelligent recommendation of logistics scheduling routes based on a large model. Background Technology

[0002] As automotive logistics transportation evolves towards parallel operations involving multiple routes, vehicles, and batches, dispatching has evolved from a simple vehicle allocation problem into a comprehensive scheduling issue that integrates route operation rhythm, driver performance capabilities, vehicle location, loading / unloading windows, return trip connections, and task status. Existing manual dispatching or simple automated dispatching systems typically rely on vehicle availability, task deadlines, route area, and administrator experience as primary criteria, enabling basic task allocation but failing to adequately consider the operational chain characteristics within automotive logistics scenarios. Automotive logistics routes generally include continuous stages such as planned journey, queuing at the factory, loading, unloading, and return trips. The actual performance difficulty of the same route varies significantly depending on the task time, vehicle location, and loading / unloading window. A driver's past experience with a particular route only indicates some route experience; it doesn't necessarily guarantee their continued ability to consistently handle tasks in the current scenario. While existing systems can maintain transportation plans, route operation data, vehicle and driver data, GPS synchronization data, and anomaly records, and can generate route familiarity assessments based on historical completion counts, historical time consumption, and anomalies, these assessments typically lean towards historical statistical results. They lack a mechanism to incorporate current task deadlines, vehicle location status, loading / unloading coordination, and backup route capabilities into scheduling decisions. This leads to dispatch results often remaining at a superficial level of matching based on "historically familiar routes" or "current vehicle availability." Especially in flexible dispatch modes where drivers independently accept orders and administrators plan allocations, the system needs to determine before a driver accepts an order whether the candidate driver / vehicle combination can arrive on time, whether loading / unloading can be smoothly coordinated, whether queuing or empty return trips will affect subsequent tasks, and whether there are available backup routes in case of main route anomalies. Existing technologies typically lack the ability to construct and extrapolate candidate solutions for the entire fulfillment process, making it difficult to compare the actual fulfillment stability of multiple driver, vehicle, and route combinations before dispatching orders. This results in recommended routes meeting static rules but experiencing problems during execution such as arrival delays, increased loading / unloading wait times, untimely backup route switching, or obstruction of subsequent task coordination. Meanwhile, when introducing large models, existing systems tend to simply use transportation plans, driver and vehicle information, and route information as input to the model as natural language prompts, allowing the model to directly provide recommendation conclusions. This lacks constraints on input fields, candidate solution selection, model output format, recommendation result verification, and task status write-back, making it difficult to verify the recommendation basis and ensuring that the model output is consistently reflected in actual order dispatch, route push, and status update actions.

[0003] Therefore, there is a need for a path intelligent recommendation method that is based on structured task profiles and candidate solutions, assisted by large-scale model performance simulation, and capable of writing the recommendation results back to the logistics scheduling business status. Summary of the Invention

[0004] The purpose of this invention is to propose an intelligent recommendation method and system for logistics scheduling routes based on a large model, in order to solve the above-mentioned problems.

[0005] To achieve the above objectives, a first aspect of the present invention provides an intelligent recommendation method for logistics scheduling routes based on a large model, the method comprising the following steps: Obtain a logistics scheduling task profile; the logistics scheduling task profile includes the current task route number, current task delivery deadline, current task status, route standard operation time, vehicle location status, vehicle availability status, driver identifier, vehicle identifier, historical completion count, historical time deviation, and historical anomaly summary; Based on the aforementioned flow scheduling task profile, a corresponding dynamic line familiarity is generated; Based on the dynamic route familiarity, and combined with the expected arrival arrangements and loading / unloading connection arrangements corresponding to the vehicle availability status, a logistics scheduling candidate plan is generated; the logistics scheduling candidate plan includes candidate drivers, candidate vehicles, candidate routes, backup routes, expected arrival arrangements, and loading / unloading connection arrangements; The input sequence, constructed based on the logistics scheduling candidate scheme, is fed into the large model to perform performance simulation according to the automotive logistics operation chain, generating logistics scheduling route recommendation results.

[0006] Furthermore, the standard operation time of the route is obtained by adding the estimated travel time, queuing time, loading time, unloading time, and return empty time; the historical time deviation is obtained by dividing the absolute deviation between the driver's historical average performance time and the standard operation time of the route by the standard operation time of the route.

[0007] Furthermore, the step of generating the corresponding dynamic route familiarity based on the flow scheduling task profile specifically involves: generating a corresponding scenario effectiveness coefficient based on the vehicle location status, historical anomaly summary, and current task deadline; and generating dynamic route familiarity based on the scenario effectiveness coefficient, combined with the historical completion count and historical time consumption deviation. The scenario effectiveness coefficient is used to represent the degree to which a driver's historical route experience is retained in the current task scenario.

[0008] Furthermore, the scenario effectiveness coefficient is generated by combining the location matching coefficient, the anomaly correction coefficient, and the delivery time compression coefficient. The location matching coefficient is obtained by mapping the vehicle location status, the anomaly correction coefficient is obtained by statistical analysis of the anomaly types in the historical anomaly summary, and the delivery time compression coefficient is calculated based on the current task delivery time requirement, the current server time, and the standard operation time of the route.

[0009] Furthermore, based on the dynamic route familiarity, and combined with the expected arrival arrangements and loading / unloading coordination arrangements corresponding to the vehicle availability status, a logistics scheduling candidate scheme is generated, specifically as follows: Based on the dynamic route familiarity, and combined with the vehicle location status, an expected arrival schedule and loading / unloading connection schedule are generated. Based on the dynamic route familiarity and the normalized task duration, the path stability and candidate scheme priority value are calculated to generate logistics scheduling candidate schemes; the path stability is used to represent the degree to which the current driver-vehicle combination can form a stable performance in this task and retain subsequent scheduling capabilities.

[0010] Furthermore, the step of generating the expected arrival schedule and loading / unloading connection schedule based on the dynamic route familiarity and the vehicle location status specifically includes: The server reads the corresponding arrival correction time from the system configuration table based on the vehicle's location status, and combines it with the estimated travel time in the standard operation time of the route to obtain the estimated arrival schedule; the estimated arrival schedule includes at least the estimated arrival time, the source of the arrival correction time, the vehicle's location status, and the valid location marker; The server generates a loading and unloading coordination schedule based on the current task plan, time, and route operation data. The task duration is calculated from the standard operation time of the line and the loading and unloading waiting time; the loading and unloading waiting time is determined based on the queuing time, loading time and unloading time in the line operation data combined with the time period configuration.

[0011] Furthermore, the priority value of the candidate scheme is calculated based on the path stability, the delivery time satisfaction coefficient, and the backup path adaptation coefficient; the delivery time satisfaction coefficient is generated by the current task delivery time requirement, the expected arrival arrangement, and the loading and unloading connection arrangement, and the backup path adaptation coefficient is jointly generated by the main path, the backup path, and the vehicle position status in the route path table.

[0012] Furthermore, the performance simulation generates recommended paths, alternative path activation suggestions, and performance reliability. After consistency verification, the logistics scheduling path recommendation results are output and the task status is updated. The reliability of performance is calculated by combining the priority value of the candidate scheme with the performance waiting offset, and the performance waiting offset is determined according to the queuing window status in the expected arrival arrangement and the loading and unloading connection arrangement. The consistency check includes checking whether the recommended driver vehicle is still available, whether the recommended route is within the candidate scheme, whether the expected loading window is consistent with the loading and unloading connection arrangement, and whether the performance reliability reaches the recommended threshold; if the check fails, the recommendation result to be confirmed is output.

[0013] Furthermore, each candidate scheme in the input sequence includes a task order number, route number, candidate driver identifier, candidate vehicle identifier, dynamic route familiarity, route stability, candidate scheme priority value, candidate route identifier, backup route identifier, expected arrival arrangement, loading and unloading connection arrangement, and task duration.

[0014] In a second aspect, the present invention provides an intelligent recommendation system for logistics scheduling routes based on a large model, the system comprising: The task profile generation module is used to obtain a logistics scheduling task profile; the logistics scheduling task profile includes the current task route number, the current task delivery deadline, the current task status, the standard operation time of the route, the vehicle location status, the vehicle availability status, the driver identifier, the vehicle identifier, the number of historical completions, the historical time deviation, and the historical anomaly summary. The dynamic familiarity generation module is used to generate corresponding dynamic line familiarity based on the flow scheduling task profile. The candidate solution generation module is used to generate logistics scheduling candidate solutions based on the dynamic route familiarity and in combination with the expected arrival arrangements and loading and unloading connection arrangements corresponding to the vehicle availability status; the logistics scheduling candidate solutions include candidate drivers, candidate vehicles, candidate routes, backup routes, expected arrival arrangements, and loading and unloading connection arrangements; The large-scale model inference and recommendation module is used to input the input sequence constructed based on the logistics scheduling candidate scheme into the large model to perform performance inference according to the automotive logistics operation chain and generate logistics scheduling route recommendation results.

[0015] The beneficial technical effects of the present invention are at least as follows: The main inventive point of this invention is that it first organizes the transportation plan, route operation, vehicle status and historical performance information in the flexible dispatch of automobile logistics into a unified logistics scheduling task profile, then generates a dynamic route familiarity based on the task profile for the current task scenario, and further expands the dynamic route familiarity to include candidate logistics scheduling schemes including candidate drivers, candidate vehicles, candidate routes, backup routes, expected arrival arrangements and loading and unloading connection arrangements. Finally, it uses a large model to perform performance simulation on the candidate schemes according to the real operation chain of arrival, queuing, loading, in transit, unloading, empty return or subsequent acceptance, and generates and executes logistics scheduling route recommendation results. This scheme elevates driver route familiarity from a purely historical evaluation to a judgment of fulfillment capability under the current task scenario. It integrates historical completion counts, historical time deviations, vehicle location status, historical anomaly summaries, and task delivery dates into the driver-vehicle combination suitability assessment. Simultaneously, it upgrades dispatch recommendation from single driver sorting to a comparison of candidate schemes in a "driver-vehicle-route-loading / unloading connection-alternative route" framework. This allows the system to identify differences in arrival arrangements, loading / unloading waiting times, task occupancy, and route switching among different candidate schemes before accepting an order. The large-scale model in this invention is used to perform fulfillment simulation and recommendation generation for pre-structured logistics scheduling candidate schemes. Its input comes from the candidate job objects formed in previous steps, and its output includes actual scheduling actions such as recommended drivers, recommended vehicles, recommended routes, alternative routes, estimated loading windows, fulfillment reliability, and task status updates. To ensure the executable output of the large model, this invention restricts the large model to run within the structured input sequence and structured output fields. The model version, input template version, number of candidate solutions, output field constraints, and result verification rules are all saved with the recommendation results. When the model output conflicts with the candidate solution fields, the recommended path does not exist, or the recommended driver vehicle is already occupied, the server does not directly execute the output, but instead falls back to the candidate solution priority value and performance reliability ranking results for verification or re-deduction.

[0016] Through the above processing, the present invention can address the problems in the prior art, such as the difficulty in reflecting current performance capability due to the familiarity with historical routes, the lack of loading and unloading operation constraints in route recommendation, the difficulty in comparing the actual performance process of candidate solutions with the dispatch results, the insufficient participation of backup route capability in recommendation, and the insufficient linkage between recommendation results and driver and administrator status. This makes logistics scheduling route recommendation more in line with the actual operation scenario of automobile logistics, which involves multiple links, high timeliness, and strong connection. Attached Figure Description

[0017] The present invention will be further described with reference to the accompanying drawings, but the embodiments in the drawings do not constitute any limitation on the present invention. For those skilled in the art, other drawings can be obtained based on the following drawings without creative effort.

[0018] Figure 1This is a flowchart of the intelligent recommendation method for logistics scheduling routes based on a large model, as described in this invention. Detailed Implementation

[0019] Embodiments of the present invention are described in detail below. Examples of these embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and should not be construed as limiting the present invention.

[0020] like Figure 1 As shown in the embodiment of the present invention, the intelligent recommendation method for logistics scheduling routes based on a large model includes: S1. Obtain a logistics scheduling task profile; the logistics scheduling task profile includes the current task route number, current task delivery deadline, current task status, route standard operation time, vehicle location status, vehicle availability status, driver identifier, vehicle identifier, historical completion count, historical time deviation, and historical anomaly summary.

[0021] Specifically, when a driver initiates an "available task" request or an administrator performs an assignment operation on a pending transportation plan, the server uses the currently pending transportation plan as the trigger object. It reads the transportation plan data, route operation data, vehicle and driver status data, and historical performance summary data that have already been maintained in the system, and organizes this data into a logistics dispatch task profile. The server first reads the task order number, route number, region, plan inclusion time, delivery deadline, and task status of the currently pending task from the transportation plan management module. The task order number is used to locate the current transportation plan, the route number is used to associate it with the route operation data, the plan inclusion time and delivery deadline are used to determine the time urgency of the task, and the task status is used to determine whether the task is in a recommendable state. All of the above fields retain the data source module, read time, and status version. If the task status has changed to assigned, canceled, or completed, the server will not generate a new recommendation profile for that task to avoid duplicate recommendations of the same transportation plan. If multiple pending tasks exist at the same time, the server filters the target transportation plans that should be included in the recommendation based on the delivery deadline and task status. For example, if a task has the route number "L-WH-032", the task status is pending assignment, and the planned inclusion time is 14:00 on the same day, then the server will continue to read route, vehicle driver, and historical performance-related data for that task.

[0022] Furthermore, the server uses the route number as an index to read the corresponding route operation data from the route information management module, including estimated travel time, queuing time, loading time, unloading time, and empty return time. This data originates from standard operation data pre-entered by the administrator on the system's route information maintenance page and is considered basic operation parameters read during order dispatch. The server records the route operation data version and effective time simultaneously. If multiple versions exist for the same route, the valid version corresponding to the current scheduled task's inclusion time is used. If a certain operation stage's duration is missing, negative, or the unit is inconsistent, the server first performs unit conversion and data verification according to the configured rules. Candidate routes that fail verification do not enter the automatic recommendation execution branch but instead enter the manual review branch. In automotive logistics transportation, route fulfillment time consists of continuous operation stages such as road travel, queuing at the factory, loading, unloading, and empty return. Therefore, the server combines these operation stages into the route's standard operation time. This calculation is based on the mathematical rule of summing similar time quantities, meaning that the total time for consecutive operation stages in the same transportation operation chain is obtained by adding the times of each stage. This invention, based on the automotive logistics scenario, treats the five stages of "driving, queuing, loading, unloading, and empty return" as fixed components, enabling the standard operating time of the route to reflect the complete fulfillment process. ; in, This indicates the standard operation time for the line, which is calculated by the server based on the line operation data maintained in the line information management module. The estimated travel time is derived from the estimated travel time field in the route information management module. The queuing time is represented by the queuing time field in the route information management module. The loading time is indicated by the loading time field in the route information management module. The unloading time is indicated by the unloading time field in the route information management module; The empty return duration is derived from the empty return duration field in the line information management module. All the above items are converted to the same measurement scale by the server before being added together. The measurement scale can be minutes, and all items should be non-negative durations. If the calculated... If the value is 0 or invalid, the server will not continue to calculate the historical time deviation and the priority value of subsequent candidate solutions, but will mark the route profile as having insufficient route operation data. Taking the "L-WH-032" route as an example, if the estimated travel time of the route is 90, the queuing time is 20, the loading time is 30, the unloading time is 25, and the return empty time is 40, then the server will obtain the standard operation time of the route as 205, and write this result into the standard operation time field of the route in the logistics scheduling task profile.

[0023] Furthermore, after the route operation data is read, the server reads vehicle and driver status data from the vehicle driver management module and the GPS synchronization module. The vehicle driver management module provides vehicle identification, driver identification, vehicle availability status, order start time, and order end time; the GPS synchronization module provides the vehicle's current location status. The server converts the GPS synchronization results into limited location statuses, such as "arrived," "waiting near the factory," "on the way back empty," and "far from the task area." The GPS synchronization results also retain the positioning time, positioning source, and positioning validity marker. If the GPS positioning is not updated within a preset valid time, the vehicle's location status is written as "location pending confirmation," and subsequent steps reduce the vehicle's candidate priority instead of directly considering it as arrived. For example, if a vehicle's GPS synchronization result shows that it is near the starting point of the target route, the vehicle's current location status is written as "waiting near the factory"; if the vehicle is still on the way back from the previous task, it is written as "on the way back empty." The server also reads historical performance summary data from the historical transportation record database and the anomaly record database using the driver identification and route number as a combined index. The historical performance summary is generated around the current route and includes historical completion counts, historical average performance time, and historical anomaly summaries. Historical completion counts are derived from the number of completed records for the same driver and the same route in completed transportation plans. Historical average performance time is derived from the driver's historical task acceptance and completion times on the same route. Historical anomaly summaries are derived from the statistical results of anomaly types archived in the anomaly record database, such as the number of loading / unloading delays, timeouts, route deviations, and anomaly reports. For abnormal performance records such as major vehicle malfunctions or traffic accidents, the server removes them as anomaly samples before generating the historical average performance time, retaining the reason for removal. If there are insufficient valid historical samples for the same driver and the same route, the historical completion count is counted based on the actual valid samples. The historical average performance time can use the effective average of all drivers on the same route or the system default value as a temporary reference, and a flag indicating insufficient historical samples is written into the profile.

[0024] The server further calculates the historical time deviation based on the historical average fulfillment time and the standard operation time for the route. This formula originates from the concept of "relative deviation" in mathematical statistics, which involves dividing the absolute difference between the actual observed value and the benchmark value by the benchmark value to obtain the degree of deviation that can be compared between different routes. In this application, the actual observed value corresponds to the driver's historical average fulfillment time, and the benchmark value corresponds to the standard operation time for the route. Due to the significant difference in the original time between long and short routes, the server uses a relative deviation method to generate the historical time deviation. ; in, Historical time deviation indicates the degree of deviation of a driver's historical performance time from the standard operating time of the route. This represents the driver's historical average fulfillment time on the current route, calculated from the historical transportation record database for the same driver and the same route; This indicates the standard operation time for the aforementioned line, calculated from the line operation data. and All use the same unit of time, and It should be a valid positive value; when When the value is the average of all drivers on the same route or the system default value, the server will... Together with the insufficient historical sample marker, this information is written into the profile, which reduces the reliability of the result when generating dynamic route familiarity in step two. Taking route "L-WH-032" as an example, the server reads that a driver's historical average fulfillment time on this route is 218, while the standard operation time for this route is 205. Therefore, the historical time deviation is... The calculated result is approximately 0.063; if another driver's historical average fulfillment time on the same route is 270, then their historical time deviation is approximately 0.317. Therefore, the server can distinguish the historical fulfillment stability of different drivers on the same route and write this difference into the logistics scheduling task profile.

[0025] Finally, after completing the above processing, the server encapsulates the transportation plan data, standard operation time for the route, vehicle and driver status data, and historical performance summary data into a logistics scheduling task profile. This profile is stored using a fixed field structure, including the current task route number, current task delivery deadline, current task status, standard operation time for the route, vehicle location status, vehicle availability status, driver identifier, vehicle identifier, historical completion count, historical time deviation, and historical anomaly summary. The profile also stores the task order number, data retrieval time, route operation data version, valid GPS location marker, number of historical samples, a list of missing fields, and the profile generation batch number for subsequent caching, sorting, and verification. For missing fields, the server marks the corresponding field as missing, enabling subsequent steps to prioritize the driver or vehicle based on the missing status. Missing fields are not assumed to be in the optimal state; when key fields involved in automatic dispatching are missing, the corresponding candidate can only enter the low-priority candidate or manual review branch. For example, if the current task to be assigned is route "L-WH-032", the system reads that the standard operation time for this route is 205 seconds, the vehicle status is available, the vehicle location status is "waiting near the factory area", the driver has completed this route 18 times in the past, the historical time deviation is 0.063 seconds, and the historical anomaly summary is "1 loading / unloading delay, 0 timeouts, 0 route deviations". The server then generates a logistics scheduling task profile corresponding to this driver and this task. If another driver's vehicle status is also available, but the historical completion count is 2 times, the historical time deviation is 0.317 seconds, and the historical anomaly summary shows multiple loading / unloading delays, the server also generates a corresponding profile for subsequent dynamic route familiarity generation steps to differentiate the processing.

[0026] S2. Generate corresponding dynamic line familiarity based on the flow scheduling task profile.

[0027] Specifically, the server uses the logistics scheduling task profile output in step one as input to generate dynamic route familiarity for the current task. The logistics scheduling task profile already includes the current task route number, current task delivery deadline, current task status, standard operating time for the route, vehicle location status, vehicle availability status, driver identifier, vehicle identifier, historical completion count, historical time deviation, and historical anomaly summary. Therefore, the server can determine the effectiveness of a driver's historical route experience under current operating conditions for the same transportation task. In flexible dispatching scenarios in automotive logistics, a driver's past completion of a route only indicates basic experience; whether the current task is suitable for that driver is also affected by the tightness of the task delivery deadline, the vehicle's current location status, historical anomaly concentration points, and historical time stability. For example, a driver's stable performance during the day, when the vehicle is already on-site, and the factory queue is short does not necessarily mean they will have the same performance capability in tasks where the vehicle is returning empty, the delivery deadline is tight, or the loading / unloading wait is long. Based on this, the server converts the static fulfillment information in the logistics scheduling task profile into dynamic route familiarity under the current task, so that subsequent logistics scheduling candidate solutions can be generated around "the driver and vehicle combination that is truly suitable to undertake this route in this task".

[0028] Further, the server first reads the current task status and vehicle availability status from the logistics scheduling task profile, and then enters the dynamic route familiarity calculation process with transportation plans in the recommendable state and available vehicles. Subsequently, the server reads the route number, driver identifier, vehicle identifier, current task deadline requirement, standard operation time for the route, vehicle location status, historical completion count, historical time deviation, and historical anomaly summary. The route number is used to confirm the target route corresponding to the dynamic route familiarity; the driver and vehicle identifiers are used to bind the calculation results to a specific driver-vehicle combination; the current task deadline requirement and standard operation time for the route are used to determine the time margin for the current task; the vehicle location status is used to determine the spatial suitability of the vehicle for the task; the historical completion count and historical time deviation are used to determine the driver's accumulated and stable historical experience; and the historical anomaly summary is used to determine whether the driver's abnormal risks on the route are concentrated in key operational aspects such as loading / unloading, timeouts, or route deviations. If the task status has changed after the profile is generated, or the vehicle has been locked by other tasks before calculation, the server marks the corresponding profile as invalid and does not continue to generate executable dynamic route familiarity.

[0029] Furthermore, the server converts the vehicle location status into location matching coefficients. This coefficient is generated by the system configuration table based on the vehicle location status mapping. The configuration table can be maintained according to the actual operating area of ​​the enterprise. For example, when the vehicle location status is "arrived," Take 1.00; when the vehicle's location status is "waiting near the factory area", Take 0.95; when the vehicle's location status is "on the way back empty", Take 0.70; when the vehicle's location status is "far from the mission area", Take 0.45. For "Location Pending Confirmation" or GPS malfunction status, The server selects values ​​based on the low-confidence level in the configuration table and retains the low-confidence marker in the calculation results. The server also converts historical anomaly summaries into anomaly correction coefficients. This coefficient is generated from the statistical results of archived exception types in the exception log database. For example, when the number of loading and unloading delays is relatively small, A value of 0.90 to 0.95 is acceptable; however, if multiple timeouts or route deviations occur, A value between 0.60 and 0.75 is acceptable. The generation rules for the anomaly correction coefficient store the anomaly type weight, statistical time range, and configuration version; if historical anomaly summaries are missing, the server will not... The default setting is no anomalies; instead, a neutral or conservative approach is adopted, and historical anomaly missing markers are written to the dynamic route familiarity record. The aforementioned location matching coefficients and anomaly correction coefficients are implemented through configuration tables or rule tables, facilitating maintenance by administrators based on the enterprise's route environment. This approach is suitable for automotive logistics scenarios because factors such as vehicle proximity to the factory, driver frequency of loading / unloading delays, and route deviations directly impact the actual reliability of the same route in the current task.

[0030] Furthermore, the server generates a delivery time compression factor based on the current task delivery time requirements and the standard operation time of the line. The server uses the current system time as a reference to calculate the available time between the current moment and the deadline, and compares this available time with the standard line operation time. When the available time is significantly longer than the standard line operation time, Take the lower value; when the available time is close to the standard operation time of the line, Take the higher value. The value ranges from 0 to 1, and the specific mapping rules are determined by the delivery time compression configuration table. When the current time exceeds the delivery time requirement, or the available time is less than the minimum time required to complete the standard operation of the route, the server marks the task as overdue or high-risk and restricts the corresponding candidate solution from entering the automatic execution branch. The setting of this coefficient comes from the classic idea of ​​"time margin" in scheduling optimization, that is, the closer the remaining available time of a task is to the standard operation time required to complete the task, the more sensitive the scheduling decision is to position deviation, historical anomalies, and time fluctuations. This application applies this idea to the automotive logistics dispatch scenario, so that the delivery time tension can participate in correcting the effectiveness of the driver's historical route experience. For example, when a task still has a sufficient margin before the delivery time, even if the vehicle is on its way back empty, it may still have the opportunity to accept the task; if the current task is only slightly longer than the standard operation time of the route before the delivery time, vehicle position deviation and historical timeout records will significantly reduce the current suitability of the driver-vehicle combination.

[0031] Furthermore, in obtaining the position matching coefficient Anomaly correction coefficient and Na period compression coefficient Afterwards, the server generates scene validity coefficients. The initial source of this formula is the multiplicative reduction concept in statistical correction, which states that when multiple independent influencing factors work together, the overall retention level can be represented by the product of normalized coefficients. This application, based on the multiplicative nature of location matching and anomaly correction, introduces a delivery-time compression term, making the reduction of the effectiveness of the current scenario more significant when the task is more urgent. ; in, The scenario validity coefficient represents the degree to which a driver's historical route experience is retained in the current task scenario. This represents the location matching coefficient, which is obtained from the vehicle location status transformation in the logistics scheduling task profile. This represents the anomaly correction coefficient, which is obtained by converting historical anomaly summaries from the logistics scheduling task profile. This represents the delivery time compression factor, generated jointly by the current task delivery time requirement, the current server time, and the standard operation time of the network. All the above variables are normalized proportional values, which can be multiplicatively combined, and are all limited to between 0 and 1 according to the configuration table before being entered into the formula. The formula... This indicates the degree of preservation of the basic scene under the combined effect of vehicle location and historical anomalies. This indicates the gap between the current scenario and the ideal performance status. This represents the amplified scenario gap under delivery time compression conditions, with the final parenthetical term used to obtain the retention ratio after the delivery time impact. Taking a driver-vehicle combination as an example, if the vehicle's location status is "waiting near the factory area," the corresponding... The value is 0.95, and the historical anomaly summary shows only minor loading and unloading delays. The value is 0.92, indicating a tight deadline for the current task. If it is 0.60, then the server calculates... The result is approximately 0.807. If the location status of the other driver's vehicle combination is "on the way back empty", the corresponding... The value is 0.70, and the historical anomaly summary contains multiple timeouts and loading / unloading delays, corresponding to... If the value is 0.65, and the current task is equally urgent, then... The result is approximately 0.306. Therefore, under the same task deadline, driver-vehicle combinations with more suitable locations and fewer historical anomalies will retain more historical route experience.

[0032] Furthermore, the server continues to generate dynamic route familiarity by combining historical completion counts and historical time deviations. Historical completion counts The "Historical Completion Count" field from the logistics dispatch task profile indicates the driver's accumulated experience on the current route; historical time deviation... The historical time deviation generated in step one represents the degree to which the driver's historical performance time deviates from the standard operating time of the route. The initial source of this formula is a combination of an experience saturation function and a relative deviation reduction concept: the experience saturation function indicates that the improvement in familiarity gradually slows down with increasing experience, while the relative deviation reduction indicates that the greater the deviation from the standard operating time, the lower the performance stability. This application combines these two with a scenario effectiveness coefficient. This approach combines historical experience, historical contract performance stability, and adaptability to the current mission scenario to ensure the final result reflects all three aspects: ; in, This represents dynamic route familiarity, indicating the driver's actual route fulfillment capability in the current task scenario; This represents the effective coefficient for the aforementioned scenario; This indicates the number of historical completions, derived from the "Number of Historical Completions" field in the logistics scheduling task profile. This indicates the historical time consumption deviation, which originates from the historical time consumption deviation field in the logistics scheduling task profile. The integer is a non-negative integer. If the driver has not completed a valid record on this route, then... If the value is 0, the dynamic line familiarity will not be misjudged as high familiarity because the experience item is 0; if If the data comes from a temporary reference value, the server retains a historical insufficient sample marker and lowers the automatic execution priority of that record in subsequent candidate solution ranking. (The formula is incomplete.) The effect of increasing the number of historical completions gradually slows down. For example, the experience gain from increasing from 1 to 2 times is significant, while the gain from increasing from 80 to 81 times is relatively weak. This term is used to convert historical time stability into a reduction term; the greater the historical time deviation, the smaller this term becomes. This is used to incorporate the impact of current delivery date, vehicle location, and historical anomalies on the final result. Each product term is normalized or proportionalized, and the final result is multiplied by 100 to form a dynamic route familiarity score that is easy to sort and threshold.

[0033] Furthermore, taking route "L-WH-032" as an example, the logistics dispatch task profile corresponding to driver A shows: historical completion count The historical time deviation is 18. The coefficient is 0.063, the vehicle's location status is "waiting near the factory area," and the historical anomaly summary only shows a small amount of loading and unloading delays. Based on the aforementioned rules, the scenario's effectiveness coefficient is obtained. If it is 0.807, then the server calculates... The result is approximately 72.0. The logistics dispatch task profile for driver B shows: historical completion count. The historical time deviation is 2. The coefficient is 0.317, the vehicle's location status is "on its way back empty," and the historical anomaly summary contains multiple records of loading / unloading delays and timeouts. The corresponding scenario's effective coefficient is... If it is 0.306, then the server calculates... The result is approximately 15.5. This calculation indicates that driver A not only has a stable track record of fulfilling contracts, but also has a more suitable current vehicle location and record of anomalies for undertaking this task; while driver B, although possessing some experience with historical routes, has a lower familiarity with the dynamic routes in the current task scenario.

[0034] Furthermore, after the server completes the dynamic line familiarity calculation, it will... The driver and vehicle identifiers are bound to the corresponding task order number, route number, driver identifier, and vehicle identifier, and written to the current task scheduling cache. If there are multiple candidate driver-vehicle combinations for the same transportation plan, the server generates dynamic route familiarity for each combination and forms the candidate ranking basis according to the dynamic route familiarity. The task scheduling cache also records the profile generation batch number, configuration table version, calculation time, and missing flags, enabling step three to identify the driver-vehicle combination. Does it still correspond to the current valid task and currently available vehicles? For example, under the current task, driver A's dynamic route familiarity is 72.0, driver B's is 15.5, and driver C's is 48.6. Then, in step three, when generating logistics scheduling candidate solutions, candidate routes, expected arrival arrangements, and loading and unloading connection arrangements will be constructed around driver A and driver C first, and driver B will be treated as a low-priority candidate.

[0035] This step ultimately outputs dynamic route familiarity. This dynamic route familiarity is generated jointly by the transportation plan, route operations, vehicle status, and historical performance information in the logistics scheduling task profile, and can convert the driver's historical route experience into a performance capability result under the current task scenario.

[0036] S3. Based on the dynamic route familiarity, and combined with the expected arrival arrangements and loading / unloading connection arrangements corresponding to the vehicle availability status, generate a logistics scheduling candidate scheme; the logistics scheduling candidate scheme includes candidate drivers, candidate vehicles, candidate routes, backup routes, expected arrival arrangements, and loading / unloading connection arrangements.

[0037] Specifically, the server generates dynamic line familiarity in step two. After that, with As the core input for the current step, and read the data related to this step. The system generates candidate logistics scheduling solutions by binding the task order number, route number, driver identifier, vehicle identifier, standard operation time for the route, vehicle location status, vehicle availability status, current task deadline requirements, and historical anomaly summaries. Dynamic route familiarity is also considered. This already reflects the driver's actual ability to fulfill the target route in the current task scenario. However, logistics scheduling route recommendation needs to further transform "whether the driver is suitable for the route" into a complete operational object: "how the driver and vehicle combination arrives, which route to take, how to connect loading and unloading, and whether there are backup routes." In the flexible dispatching scenario of automotive logistics, the same task may have multiple driver and vehicle combinations. Different combinations are simultaneously affected by the vehicle's current location, factory queuing window, loading and unloading rhythm, empty return occupancy, and route switching capability. Therefore, the server generates a preliminary candidate solution based on each dynamic route familiarity record, and then expands the preliminary candidate solution into a logistics scheduling candidate solution that can be used for large-scale model performance simulation. The task status, vehicle availability status, route table, and loading and unloading time period configuration used in this step are all based on the latest valid version at the time of generating the candidate solution. If it is found that the vehicle or task cached in step two has expired, the corresponding record will not be included in the candidate solution set.

[0038] Furthermore, the server first reads all dynamic route familiarity records corresponding to the current task route number from the task scheduling cache. Each record contains... The server determines the driver-vehicle combinations for candidate solution generation based on vehicle availability and order acceptance time range. Vehicle availability is derived from the logistics dispatch task profile, and order acceptance time range is derived from the order start and end times maintained in the vehicle and driver management module. For driver-vehicle combinations within the order acceptance time range and with available vehicle status, the server reads the vehicle location status, standard operation time for the route, and current task deadline requirements, and generates preliminary candidate solutions. The server also performs occupancy and locking checks on candidate vehicles. If a vehicle is already occupied by other incomplete tasks, or the driver has withdrawn their idle request, no executable candidate solution is generated for that combination; only unfiltered records are retained. For example, for route "L-WH-032," there are currently three driver-vehicle combinations: Driver A, Driver B, and Driver C. Step two yields their dynamic route familiarity scores of 72.0, 15.5, and 48.6, respectively. The server then generates preliminary candidate solutions for each of these three combinations and continues to calculate their expected arrival arrangements, loading / unloading coordination arrangements, and route stability.

[0039] Furthermore, the server generates an estimated arrival schedule based on the vehicle's location status. The vehicle location status is derived from the logistics scheduling task profile and has been converted from GPS synchronization results into limited states such as "Arrived," "Waiting Near the Factory," "On the Way Back," and "Far From the Task Area." Based on the vehicle's location status, the server reads the corresponding arrival correction time from the system configuration table and combines it with the estimated travel time from the standard operation time for the route to obtain the estimated arrival schedule. For example, when the vehicle's location status is "Arrived," the arrival correction time is relatively short; when the vehicle's location status is "Waiting Near the Factory," the arrival correction time is slightly longer than that of the "Arrived" status; when the vehicle's location status is "On the Way Back," the server reads the average switching time of return missions in the same area from the historical return record table and adds this average switching time to the estimated arrival schedule; when the vehicle's location status is "Far From the Task Area," the server sets the estimated arrival schedule for that vehicle combination to a later time slot. The expected arrival schedule should include at least the estimated arrival time, the source of the arrival correction time, the vehicle's location status, and a valid location marker. If the location is invalid or there are insufficient empty return records, the expected arrival schedule should be marked as low-reliability, and it should not be used as a basis for high-reliability arrival in subsequent candidate scheme priority values ​​and large-scale model extrapolations. This expected arrival schedule is used to describe the vehicle's ability to switch from its current operating state to the target loading point, and is suitable for representing the actual operational characteristics of vehicle travel to and from the factory, queuing, and close task connections in automotive logistics.

[0040] Furthermore, after obtaining the expected arrival schedule, the server generates a loading and unloading coordination arrangement based on the current task plan inclusion time and route operation data. The server reads the queuing time, loading time, and unloading time corresponding to the target route from the route information management module, and determines the corresponding low queuing period, normal queuing period, or high queuing period based on the time period of the planned inclusion time, calling the route operation time period configuration table. For high queuing periods, the server increases the loading and unloading waiting time; for normal queuing periods, the server adopts the standard queuing time maintained by the route information management module; for low queuing periods, the server reduces the loading and unloading waiting time. After the loading and unloading waiting time is written into the candidate scheme prototype, it forms the loading and unloading coordination arrangement together with the expected arrival schedule. The loading and unloading coordination arrangement also records the time period configuration version, queuing window type, expected loading start time, and expected unloading completion time; when the expected arrival time is later than the loading window or cannot meet the delivery time requirements, the candidate scheme can be retained for comparison, but it is marked as a low-priority candidate and will not be included in the automatic recommended execution results. For example, although vehicle A is closer to the loading point, its expected arrival time falls within the high queuing period; vehicle C is slightly farther away, but its expected arrival time falls within the normal queuing period. Therefore, the two vehicles will exhibit different work occupancy situations in subsequent path stability.

[0041] Furthermore, the server generates path stability. Path stability represents the degree to which the current driver-vehicle combination achieves stable fulfillment in this task and retains subsequent scheduling capacity. This calculation originates from the resource occupancy reduction concept in scheduling optimization, namely, the longer a vehicle is occupied by the current task, the lower its scheduling capacity for the subsequent task pool; based on this concept, this application uses the dynamic route familiarity output in step two... Combined with the current task duration, the route stability reflects both "the driver-vehicle combination is suitable for the current route" and "the current task's vehicle resource usage is controllable." The server first generates the task duration based on the standard operation time of the route and the loading / unloading waiting time. The standard operation time for the route comes from the logistics scheduling task profile in step one, while the loading and unloading waiting time comes from the loading and unloading connection arrangement generated in this step. The task duration is then normalized so that it is used in the calculation as a proportion. ; in, Path stability is used to indicate the degree to which the current driver-vehicle combination can stably fulfill its obligations in this task and retain the ability to be dispatched in the future. The dynamic route familiarity generated in step two comes from records in the task scheduling cache that are bound to the current task order number, route number, driver identifier, and vehicle identifier. This represents the normalized task duration, which is obtained by combining the standard operation time of the line with the loading / unloading waiting time generated in this step. The normalization metric is determined by the system configuration table and saved with the candidate schemes; if the standard operation time of the line or the loading and unloading waiting time is missing, a high-reliability metric will not be generated. The corresponding candidate solution will then enter the low-priority or manually reviewed branch. In this formula, This is a resource consumption reduction item; the longer the task takes, the smaller this reduction item becomes. Higher dynamic route familiarity and lower task duration result in higher route stability. Taking route "L-WH-032" as an example, driver A's dynamic route familiarity... The time taken is 72.0, the standard operation time for the line is 205, the current loading and unloading waiting time is 25, the server receives the task and the time taken is 230, which is converted according to the normalized scale configured in the system. ,but The calculated result is approximately 21.8; Driver C's dynamic route familiarity... The time taken for the standard operation of the line is 48.6, the time taken for the line to complete the operation is 205, the current high queue period corresponds to a loading and unloading waiting time of 60, and the time taken for the server to receive the task is 265, which is normalized to... ,but The calculated result is approximately 13.3. Therefore, the server can integrate dynamic route familiarity and the current task's resource consumption of vehicles into the candidate solution ranking.

[0042] Furthermore, the server obtains path stability. Then, priority values ​​for candidate solutions are generated. The priority value of candidate solutions is used to determine which candidate solutions enter the large model performance simulation set in step four. This calculation is derived from the weighted ranking idea in multi-objective scheduling, which combines multiple scheduling objectives after converting them into a unified scale; this application applies it to the flexible dispatch scenario of automotive logistics, adding a delivery time satisfaction coefficient to the path stability. Adaptability coefficient of backup path This ensures that candidate solutions simultaneously reflect performance stability, on-time fulfillment, and path switching capabilities. The server generates the solution based on the current task's on-time requirements, expected arrival arrangements, and loading / unloading coordination arrangements. When the expected arrival arrangements and loading / unloading coordination can meet the delivery deadline and allow for operational leeway, Take the higher value; when the expected arrival schedule is close to the delivery deadline or the loading and unloading coordination may reduce the operational margin. Reduce; when the expected arrival arrangements and loading / unloading coordination arrangements can no longer meet the delivery time requirements, the server marks the candidate solution as a low-priority candidate. The values ​​are determined by the delivery period meeting the configuration table, and its configuration version is saved. The server generates the configuration based on the route path table. The route table originates from the route information management module and records the primary and backup routes of the target route. When a usable backup route exists for the target route, and the vehicle's current location is suitable for switching to the backup route, Take the higher value; when the alternative route is heavily occupied or the vehicle's current location is not suitable for switching. Reduced; when alternative paths are unavailable or the path table is missing. The lowest value is taken and added as a missing alternative path marker. The server then calculates the priority of the candidate solutions based on this. ; in, This indicates the priority value of the candidate solution, used to rank multiple logistics scheduling candidate solutions for the same task; This represents path stability, calculated using the aforementioned formula; This represents the delivery deadline satisfaction factor, which is generated based on the current task delivery deadline requirements, expected arrival arrangements, and loading / unloading coordination arrangements. This represents the alternative path adaptation coefficient, which is generated jointly by the main path, alternative path, and vehicle location status in the route path table. First, demonstrate the basic stability of the candidate solutions. This is used to increase the priority of solutions that meet delivery deadlines and have operational margins when entering the simulation set. This is used to give higher priority to schemes that have the conditions for switching to alternative paths. Taking driver A as an example, their path stability... The value is 21.8. It is anticipated that the on-site arrangements and loading / unloading coordination will meet the delivery schedule requirements and allow for operational leeway. The value is 0.30, and there is a suitable alternative path for the target route at its current location. If it is 0.15, then The calculated result is approximately 32.6; the path stability of driver C. The value is 13.3. Although the on-site arrangements and loading / unloading coordination are expected to meet the delivery deadline, the operational margin is relatively small. The value is 0.10, indicating that the conditions for switching to a backup path are generally normal. If it is 0.05, then The calculated result is approximately 15.4. Based on this, the server ranks driver A's corresponding scheme before driver C and writes the ranking result into the task scheduling cache.

[0043] Furthermore, after calculating the priority values ​​of the candidate solutions, the server proceeds according to... A predetermined number of logistics scheduling candidate solutions are selected from high to low to form the input set for the large model derivation in step four. Each logistics scheduling candidate solution adopts a fixed field structure, including task order number, route number, candidate driver identifier, candidate vehicle identifier, and dynamic route familiarity. Path stability Candidate solution priority value Candidate route identifier, backup route identifier, estimated arrival schedule, loading and unloading coordination schedule, and normalized task duration. The candidate route identifier and backup route identifier are derived from the route route table in the route information management module; the estimated arrival schedule is derived from the vehicle location status and arrival correction time; the loading and unloading connection schedule is derived from the queuing time, loading time, and unloading time in the route operation data; and the normalized task occupancy time is also included. The data is derived from the standard operation time and loading / unloading waiting time of the route. Candidate solutions also store vehicle occupancy verification results, low-confidence markers, configuration versions, and candidate solution generation times. When the same driver and vehicle combination corresponds to multiple pending tasks, the server performs conflict checks based on task delivery dates and candidate solution priority values ​​to prevent the same vehicle from being occupied by multiple recommended results simultaneously. Taking the "L-WH-032" route as an example, the server generates candidate solution one as follows: Driver A, Vehicle A1, Main Route P1, Backup Route P2, Dynamic Route Familiarity 72.0, Route Stability 21.8, Candidate Solution Priority 32.6, Estimated Arrival Time 15:20, Loading / Unloading Connection Arrangement is a Normal Queue Window; Candidate solution two is as follows: Driver C, Vehicle C1, Main Route P1, Backup Route P3, Dynamic Route Familiarity 48.6, Route Stability 13.3, Candidate Solution Priority 15.4, Estimated Arrival Time 15:45, Loading / Unloading Connection Arrangement is a High Queue Window. The server writes these candidate solutions into the task scheduling cache for step four to perform intelligent recommendation of logistics scheduling routes based on the large model.

[0044] This step ultimately outputs a candidate logistics scheduling solution. This candidate solution is derived from the dynamic route familiarity data output in step two, and is combined with vehicle location status, loading / unloading connection arrangements, route stability, delivery time satisfaction coefficient, and alternative route suitability coefficient to form an executable logistics operation object. Subsequent steps use the candidate logistics scheduling solution as input, and the large model simulates the fulfillment process of each candidate solution, generating the final recommended logistics scheduling route.

[0045] S4. Input the input sequence constructed based on the logistics scheduling candidate scheme into the large model to perform performance simulation according to the automotive logistics operation chain, and generate logistics scheduling path recommendation results.

[0046] Specifically, after the server outputs logistics scheduling candidate solutions in step three, it uses these candidate solutions as input for the current step to execute intelligent logistics scheduling route recommendation based on a large model. Each logistics scheduling candidate solution includes a task order number, route number, candidate driver identifier, candidate vehicle identifier, and dynamic route familiarity. Path stability Candidate solution priority value Candidate route identifier, backup route identifier, estimated arrival schedule, loading and unloading coordination schedule, and task duration. The server prioritizes candidate solutions. The system reads a predetermined number of candidate solutions from high to low and organizes these solutions into a structured input sequence for a large model. This allows the large model to perform performance simulations based on multiple candidate drivers, vehicles, and routes under the same task. The structured input sequence uses the task order number as the simulation object index, the route number as the route and operation rule index, the candidate driver and vehicle identifiers as the receiving object index, and the candidate route and backup route identifiers as the route execution object index, along with dynamic route familiarity. Path stability Candidate solution priority value Expected arrival arrangements, loading and unloading coordination arrangements, and task duration As parameters for performance simulation, the structured input sequence undergoes field integrity checks, route existence checks, and vehicle occupancy checks before being fed into the large model. Candidate solutions that fail the checks are not recommended as executable options. Therefore, the input read by the large model is the complete logistics operation object formed in step three, rather than the scattered raw business data.

[0047] Furthermore, the large model deployed on the server adopts a Transformer-based sequence generation structure, including an input embedding layer, a multi-head self-attention layer, a feedforward network layer, a temporal position encoding layer, and a structured output layer. The input embedding layer converts candidate driver identifiers, candidate vehicle identifiers, route numbers, candidate route identifiers, and alternative route identifiers into discrete embedding vectors, dynamically representing route familiarity. Path stability Candidate solution priority value Normalized task duration The expected arrival schedule and loading / unloading connection schedule are converted into continuous numerical vectors and then into operation time series vectors. A multi-head self-attention layer simultaneously focuses on the relationships between drivers, vehicles, routes, and loading / unloading operations within each candidate solution. For example, if a candidate driver has high familiarity with the dynamic route but their expected arrival time falls within a high queuing window, the model can simultaneously focus on both "driver familiarity with the route" and "increased loading / unloading waiting time." A feedforward network layer performs a non-linear mapping on the combined features output by the attention layer to obtain the intermediate states of each candidate solution at different fulfillment stages. A temporal location encoding layer expands the candidate solutions into a sequence of "arrival, queuing, loading, en route, unloading, empty return or subsequent pick-up" according to the automotive logistics operation chain, ensuring that the model's deduction results conform to the sequence of transportation operations. The structured output layer outputs the recommended route, alternative route activation suggestion, expected loading window, and recommendation reason for each candidate solution, and writes it back to the server in a fixed field format. The model version, parameter version, input field template, and output field template of the large model are all saved with this recommendation batch; the model only compares and infers within the candidate solution set, and does not generate driver, vehicle, or route identifiers outside the candidate set, thereby avoiding the recommendation results from deviating from the executable candidate objects formed in step three.

[0048] Furthermore, before inputting the large model, the server generates a performance waiting offset based on the expected arrival arrangements and loading / unloading coordination arrangements in the candidate schemes. The fulfillment wait offset represents the wait offset caused by the factory's queuing window, loading window, and loading / unloading connection status after the vehicle's expected arrival. The server reads the expected arrival schedule from the candidate schemes and reads the queuing window status in the loading / unloading connection schedule; when the expected arrival schedule falls into a normal queuing window, Take the lower value; when the expected attendance falls within a high queue window, Take the higher value; when the expected arrival is earlier than the loading window and can enter the waiting queue. Generate according to the waiting queue occupancy rules. The parameters are determined by the work period configuration table, queue window status, and expected arrival schedule. Values ​​are taken using a normalized ratio, and configuration versions are saved. When the loading / unloading connection arrangement is of low reliability or the expected arrival schedule is missing, the server will... The candidate solution is marked as low confidence and its direct inclusion in the automatic execution recommendation is restricted. This processing is derived from the waiting propagation concept in queuing theory, where waiting in one operation propagates to subsequent operations. This application applies this concept to the automotive logistics loading and unloading scenario, combining the arrival arrangements in the candidate solutions with the queuing status in the factory area to form a waiting offset input that can be used for large-scale model extrapolation and for the server to calculate the fulfillment reliability. For example, if candidate solution one has an estimated arrival time of 15:20 and the loading and unloading connection arrangement is a normal queuing window, the server obtains the waiting offset input based on the operation time slot configuration table. Candidate Option 2 is expected to arrive at 15:45. The loading and unloading arrangement is a high-queue window, and the server obtains the configuration based on the same configuration table. The values ​​above are all normalized proportional values, used to represent the degree of waiting offset.

[0049] Furthermore, the server prioritizes candidate solutions. and performance wait offset Generate performance reliability This formula originates from the reliability engineering concept of "capability value reduced by risk," meaning that the higher the basic capability and the smaller the risk reduction, the higher the final reliability. This application prioritizes candidate solutions. As the basic capability value of candidate solutions before entering the performance simulation, the performance waiting offset will be used. As a risk reduction factor for loading / unloading waiting and queuing propagation, the fulfillment reliability used for the final recommendation ranking is obtained: ; in, It represents the reliability of performance, used to indicate the degree of stability of the final performance of candidate solutions in the real logistics operation chain; This indicates that the priority value of the candidate solution generated in step three comes from the priority value field of the candidate solution in the logistics scheduling candidate solution; This represents the performance waiting offset, generated from the expected arrival arrangements and loading / unloading coordination arrangements in the candidate schemes. In order to wait for the risk reduction item, The larger the value, the lower the reliability of the candidate solution in fulfilling its obligations. higher and The lower the value, the more suitable the candidate solution is as the final recommended solution. The performance deduction sequence output by the large model is used together to generate recommendation results. If the recommendation scheme output by the large model is consistent with... If the ranking difference exceeds a preset review threshold, the server marks the result as requiring administrator confirmation instead of automatically executing the review. Taking the "L-WH-032" line as an example, the priority value of candidate scheme one... Performance waiting offset Then the server calculates The result is approximately 26.08; the priority value of candidate solution two is... Performance waiting offset Then the server calculates The result is approximately 8.11. This calculation process allows candidate solutions for the same task to be compared on a uniform scale and makes the diffusion of loading and unloading wait time have a clear impact on the final recommendation result.

[0050] Furthermore, the server will include the task order number, route number, candidate driver identifier, candidate vehicle identifier, candidate route identifier, alternative route identifier, and dynamic route familiarity for each candidate solution. Path stability Candidate solution priority value Normalized task duration Expected arrival arrangements, loading and unloading coordination arrangements, and performance delays. and performance reliability Input the large model. The large model generates a performance simulation sequence according to the order of arrival, queuing, loading, en route, unloading, empty return or subsequent acceptance. For each candidate scenario, the large model first generates the expected arrival node and determines whether the node matches the loading and unloading connection arrangement; then it generates queuing nodes, combined with the performance waiting offset. The impact of queuing on loading start time is simulated; then loading nodes, in-transit nodes, and unloading nodes are generated, and the path execution process is simulated by combining the primary path corresponding to the candidate path identifier and the backup path corresponding to the backup path identifier; finally, empty return or subsequent takeover nodes are generated, combined with the normalized task duration. The model simulates the vehicle's availability status after completing its current task. The performance simulation sequence is output in the form of structured nodes. Each node includes a node type, estimated time, referenced candidate solution fields, and a risk marker. Recommendation reasons can only reference these structured fields and are not allowed to generate free text facts inconsistent with system records. For example, if the estimated arrival time for candidate solution one is 15:20, the loading and unloading arrangement is a standard queuing window, the main path is P1, and the backup path is P2, then the large model can output a performance simulation sequence of "arrival at 15:20, loading begins at 15:35, main path P1 is executed, backup path P2 is reserved, and the vehicle is expected to return empty after unloading."

[0051] Furthermore, when generating the performance simulation sequence, the large model performs path switching judgments on the primary and backup paths. The server reads the primary path information corresponding to the candidate path identifier and the backup path information corresponding to the backup path identifier from the route table; the large model reads the historical congestion records of the primary path, the detour occupancy of the backup path, and the vehicle location status in the candidate schemes to generate recommended paths and backup path activation suggestions. When the primary path has relatively stable historical traffic records in the current time period, the large model keeps the primary path as the recommended path; when the primary path has high historical congestion records and the backup path has low detour occupancy, the large model writes the backup path into the backup path activation suggestion. After the path switching suggestion is output, the server verifies again whether the recommended path and backup path exist in the route table, whether they are in an activated state, and whether they match the current vehicle location status; path switching suggestions that fail the verification do not enter the GPS push process, but enter the administrator confirmation or re-simulation process. This simulation process is adapted to the business model of long-term maintenance of fixed primary routes and fixed backup routes in automotive logistics, so that the recommendation results can be implemented in the system's existing route table and GPS push process.

[0052] Furthermore, the server determines the reliability of each candidate solution's performance. Based on the fulfillment projection sequence output by the large model, a logistics scheduling route recommendation result is generated. The logistics scheduling route recommendation result includes at least the task order number, recommended driver identifier, recommended vehicle identifier, recommended route identifier, alternative route identifier, estimated arrival time, estimated loading window, and fulfillment reliability. The recommendation reasons are generated by the structured output layer of the large model and are limited to combinations of logistics scheduling fields, such as "high familiarity with dynamic routes, estimated arrival time matches loading and unloading windows, stable main route, and switchable backup routes." Before generating the recommendation results, the server performs consistency checks, including whether the recommended driver and vehicle are still available, whether the recommended route is within the candidate options, whether the estimated loading window matches the loading and unloading connection arrangements, and whether the fulfillment reliability reaches the recommendation threshold. If the consistency check fails, the server outputs a recommendation result pending confirmation instead of automatically dispatching orders. Taking the "L-WH-032" route as an example, if candidate option one corresponds to driver A, vehicle A1, recommended main route P1, backup route P2, estimated arrival time 15:20, estimated loading window 15:35 to 15:50, and fulfillment reliability 26.08, and candidate option two corresponds to driver C, vehicle C1, and fulfillment reliability 8.11, then the server selects candidate option one as the recommended logistics scheduling route.

[0053] The server then executes the recommended results. For scenarios triggered by the administrator, the server pushes the logistics scheduling route recommendation results to the administrator. After the administrator confirms, the recommended driver and vehicle are bound together, and the transportation plan status is updated to "assigned." For scenarios where the driver initiates an idle order request, the server pushes the recommended task, recommended route, and expected arrival arrangement to the driver. After the driver confirms acceptance, the server updates the task status to "assigned" and synchronizes the recommended route identifier to the GPS system. If the task status, vehicle availability, GPS location, or loading / unloading window changes before administrator or driver confirmation, the server pauses the execution of the recommendation result and returns to step one or step three to regenerate the profile or candidate solution. After the vehicle starts executing the task, the system updates the task status to "in operation" based on the driver's operation and GPS synchronization results. After the task is completed, the system updates the task status to "completed" and writes the actual fulfillment time, actual route usage, and actual queuing status into the historical transportation record database and the exception record database. Thus, the recommended results move from large-scale model fulfillment deduction to the actual order dispatch, route push, and transportation plan status update process.

[0054] This step ultimately outputs the recommended logistics scheduling route and the updated task status. The recommended logistics scheduling route is obtained from the logistics scheduling candidate schemes output in step three through large-scale model performance simulation. The updated task status is triggered jointly by driver confirmation, administrator confirmation, and GPS synchronization. The recommendation result also saves the model version, input template version, output template version, candidate scheme number, performance simulation sequence, and performance reliability. The consistency verification results and execution status facilitate subsequent review and recommendation. When the system generates a logistics scheduling task profile again, the new actual performance data will be included in step one as a historical performance summary, enabling subsequent dynamic route familiarity and candidate solution generation to continue to be calculated based on real operational results.

[0055] This invention also provides an intelligent recommendation system for logistics scheduling routes based on a large model, the system comprising: The task profile generation module is used to obtain a logistics scheduling task profile; the logistics scheduling task profile includes the current task route number, the current task delivery deadline, the current task status, the standard operation time of the route, the vehicle location status, the vehicle availability status, the driver identifier, the vehicle identifier, the number of historical completions, the historical time deviation, and the historical anomaly summary. The dynamic familiarity generation module is used to generate corresponding dynamic line familiarity based on the flow scheduling task profile. The candidate solution generation module is used to generate logistics scheduling candidate solutions based on the dynamic route familiarity and in combination with the expected arrival arrangements and loading and unloading connection arrangements corresponding to the vehicle availability status; the logistics scheduling candidate solutions include candidate drivers, candidate vehicles, candidate routes, backup routes, expected arrival arrangements, and loading and unloading connection arrangements; The large-scale model inference and recommendation module is used to input the input sequence constructed based on the logistics scheduling candidate scheme into the large model to perform performance inference according to the automotive logistics operation chain and generate logistics scheduling route recommendation results.

[0056] Those skilled in the art will understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0057] In the embodiments provided in this application, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection of apparatuses or units may be electrical, mechanical, or other forms.

[0058] If the aforementioned functions are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0059] Although embodiments of the invention have been shown and described, those skilled in the art will understand that various changes, modifications, substitutions and variations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the claims and their equivalents.

Claims

1. A method for intelligent recommendation of logistics scheduling routes based on a large model, characterized in that, The method includes the following steps: Obtain a logistics scheduling task profile; the logistics scheduling task profile includes the current task route number, current task delivery deadline, current task status, route standard operation time, vehicle location status, vehicle availability status, driver identifier, vehicle identifier, historical completion count, historical time deviation, and historical anomaly summary; Based on the aforementioned flow scheduling task profile, a corresponding dynamic line familiarity is generated; Based on the dynamic route familiarity, and combined with the expected arrival arrangements and loading / unloading connection arrangements corresponding to the vehicle availability status, a logistics scheduling candidate plan is generated; the logistics scheduling candidate plan includes candidate drivers, candidate vehicles, candidate routes, backup routes, expected arrival arrangements, and loading / unloading connection arrangements; The input sequence, constructed based on the logistics scheduling candidate scheme, is fed into the large model to perform performance simulation according to the automotive logistics operation chain, generating logistics scheduling route recommendation results.

2. The intelligent recommendation method for logistics scheduling routes based on a large model according to claim 1, characterized in that, The standard operating time for the route is obtained by adding the estimated travel time, queuing time, loading time, unloading time, and return empty time; the historical time deviation is obtained by dividing the absolute deviation between the driver's historical average performance time and the standard operating time of the route by the standard operating time of the route.

3. The intelligent recommendation method for logistics scheduling routes based on a large model according to claim 1, characterized in that, The process of generating dynamic route familiarity based on the logistics scheduling task profile specifically involves: generating a corresponding scenario effectiveness coefficient based on the vehicle location status, historical anomaly summary, and current task deadline; and generating dynamic route familiarity based on the scenario effectiveness coefficient, combined with the historical completion count and historical time consumption deviation. The scenario effectiveness coefficient is used to represent the degree to which a driver's historical route experience is retained in the current task scenario.

4. The intelligent recommendation method for logistics scheduling routes based on a large model according to claim 3, characterized in that, The scenario effectiveness coefficient is generated by combining the location matching coefficient, the anomaly correction coefficient, and the delivery time compression coefficient. The location matching coefficient is obtained by mapping the vehicle location status, the anomaly correction coefficient is obtained by statistical analysis of the anomaly types in the historical anomaly summary, and the delivery time compression coefficient is calculated based on the current task delivery time requirement, the current server time, and the standard operation time of the route.

5. The intelligent recommendation method for logistics scheduling routes based on a large model according to claim 1, characterized in that, Based on the dynamic route familiarity, and combined with the expected arrival arrangements and loading / unloading coordination arrangements corresponding to the vehicle availability status, a logistics scheduling candidate scheme is generated, specifically as follows: Based on the dynamic route familiarity, and combined with the vehicle location status, an expected arrival schedule and loading / unloading connection schedule are generated. Based on the dynamic route familiarity and the normalized task duration, the path stability and candidate scheme priority value are calculated to generate logistics scheduling candidate schemes; the path stability is used to represent the degree to which the current driver-vehicle combination can form a stable performance in this task and retain subsequent scheduling capabilities.

6. The intelligent recommendation method for logistics scheduling routes based on a large model according to claim 5, characterized in that, The process of generating the expected arrival schedule and loading / unloading connection schedule based on the dynamic route familiarity and the vehicle location status is as follows: The server reads the corresponding arrival correction time from the system configuration table based on the vehicle's location status, and combines it with the estimated travel time in the standard operation time of the route to obtain the estimated arrival schedule; the estimated arrival schedule includes at least the estimated arrival time, the source of the arrival correction time, the vehicle's location status, and the valid location marker; The server generates a loading and unloading coordination schedule based on the current task plan, time, and route operation data. The duration of the task is calculated from the standard operation time of the line and the waiting time for loading and unloading. The loading and unloading waiting time is determined based on the queuing time, loading time, and unloading time in the line operation data, combined with the time period configuration.

7. The intelligent recommendation method for logistics scheduling routes based on a large model according to claim 5, characterized in that, The priority value of the candidate scheme is calculated based on the path stability, the delivery time satisfaction coefficient, and the backup path adaptation coefficient. The delivery time satisfaction coefficient is generated by the current task delivery time requirement, the expected arrival arrangement, and the loading and unloading connection arrangement. The backup path adaptation coefficient is jointly generated by the main path, the backup path, and the vehicle position status in the route path table.

8. The intelligent recommendation method for logistics scheduling routes based on a large model according to claim 1, characterized in that, The performance simulation generates recommended paths, backup path activation suggestions, and performance reliability. After consistency verification, the logistics scheduling path recommendation results are output and the task status is updated. The reliability of performance is calculated by combining the priority value of the candidate scheme with the performance waiting offset, and the performance waiting offset is determined according to the queuing window status in the expected arrival arrangement and the loading and unloading connection arrangement. The consistency check includes checking whether the recommended driver vehicle is still available, whether the recommended route is within the candidate scheme, whether the expected loading window is consistent with the loading and unloading connection arrangement, and whether the performance reliability reaches the recommended threshold; if the check fails, the recommendation result to be confirmed is output.

9. The intelligent recommendation method for logistics scheduling routes based on a large model according to claim 1, characterized in that, Each candidate scheme in the input sequence includes a task order number, route number, candidate driver identifier, candidate vehicle identifier, dynamic route familiarity, route stability, candidate scheme priority value, candidate route identifier, backup route identifier, expected arrival arrangement, loading and unloading connection arrangement, and task duration.

10. A logistics scheduling route intelligent recommendation system based on a large model, characterized in that: The system includes: The task profile generation module is used to obtain a logistics scheduling task profile; the logistics scheduling task profile includes the current task route number, the current task delivery deadline, the current task status, the standard operation time of the route, the vehicle location status, the vehicle availability status, the driver identifier, the vehicle identifier, the number of historical completions, the historical time deviation, and the historical anomaly summary. The dynamic familiarity generation module is used to generate corresponding dynamic line familiarity based on the flow scheduling task profile. The candidate solution generation module is used to generate logistics scheduling candidate solutions based on the dynamic route familiarity and in combination with the expected arrival arrangements and loading and unloading connection arrangements corresponding to the vehicle availability status; the logistics scheduling candidate solutions include candidate drivers, candidate vehicles, candidate routes, backup routes, expected arrival arrangements, and loading and unloading connection arrangements; The large-scale model inference and recommendation module is used to input the input sequence constructed based on the logistics scheduling candidate scheme into the large model to perform performance inference according to the automotive logistics operation chain and generate logistics scheduling route recommendation results.

Images