A short-distance transportation order intelligent scheduling method and system based on template fingerprint grouping feedback

Abstract

The present application relates to the technical field of short-distance transportation intelligent scheduling, logistics information processing and computer application, in particular to a short-distance transportation order intelligent scheduling method and system based on template fingerprint grouping feedback and a storage medium. The method establishes a route responsibility semantic index library, generates a template fingerprint for an imported order, calls a location synonym sub-dictionary to normalize the starting and ending points, determines a target route in combination with a responsibility inverted index, generates a restricted candidate resource pool based on a responsibility vehicle set, a responsibility driver set and a load constraint, executes whole-vehicle direct dispatch, compatible carpooling, return vehicle reuse and overload splitting, and updates the location synonym sub-dictionary and related indexes locally according to the manual route change result. The scheme improves the route recognition accuracy, scheduling efficiency and resource utilization rate.

Description

Technical Field

[0001] This invention relates to the fields of intelligent scheduling of short-distance transportation, logistics information processing, and computer application technology, and in particular to an intelligent scheduling method and system for short-distance transportation orders based on template fingerprint grouping feedback. Background Technology

[0002] With the continuous development of supply chain collaboration in manufacturing enterprises, internal logistics organization within industrial parks, and warehousing and transshipment businesses, short-distance transportation scenarios such as raw material entry, semi-finished product transfer, finished product short-distance transport, and inter-warehouse allocation are increasingly common. Compared with long-distance trunk transportation, short-distance transportation typically features more order batches, larger variations in single-ticket weight, concentrated delivery time requirements, and relatively fixed but inconsistent writing methods for transportation origin and destination. Especially in internal enterprise application scenarios, transportation needs are often submitted by multiple departments such as purchasing, warehousing, production, and planning through spreadsheets or other text formats. Different departments have different habits in filling in the origin, destination, material category, and route name, which can easily lead to abbreviations, aliases, colloquial names, historical names, or template differences, resulting in the same location being described in multiple text formats. If the main reliance is still on manual location identification, route judgment, vehicle assignment, and driver arrangement, not only is the dispatching efficiency low, but it is also prone to problems such as dispatching errors, high empty running rates, unreasonable carpooling, idle return resources, and untraceable abnormal adjustments, which can hardly meet the requirements of modern short-distance transportation management for real-time performance, accuracy, and traceability. The development of existing logistics scheduling technologies also shows that, with the expansion of business scale and the increase in process complexity, it is difficult to balance response speed and processing accuracy by relying solely on human experience for transportation arrangements.

[0003] In the prior art, Chinese patent CN105512747B discloses a logistics intelligent optimization scheduling system. The system proposed in this document includes a basic data module, an intelligent scheduling module, and a Kanban management module. The basic data module further includes a route management unit, a vehicle type management unit, an address management unit, a carpooling rule unit, a scheme design unit, a capacity management unit, a data dictionary unit, and a region management unit. The intelligent scheduling module mainly realizes the complete scheduling process from order extraction to scheduling calculation, result review, and writing back to the transportation management system. This document also explicitly mentions that routes are the basic object for order extraction and scheduling; the address management unit is used to uniformly maintain address information; the carpooling rule unit is used to maintain rules for cities where carpooling is allowed and rules for prohibiting carpooling; and the capacity management unit is used to manage vehicle capacity status. Therefore, this document has attempted to improve the automation level of logistics transportation from the perspectives of basic data maintenance, transportation resource management, and automatic scheduling calculation, which can reduce inventory, labor, and transportation costs to a certain extent and improve on-time delivery capabilities. However, the overall content of this patent still focuses more on route maintenance, address maintenance, capacity management, and carpooling rule management in a standardized transportation management system environment. Based on its publicly available content, its address management primarily emphasizes the unified maintenance of addresses and the consistency of address identification during order extraction. Carpooling rules are also largely reflected in route-level rule maintenance and scheduling execution. However, the document does not provide specific mechanisms to address common issues in short-distance transportation within enterprises, such as importing multiple template orders, differences in templated aliases for the same location, the creation of local dictionary feedback based on historical manual route changes, and the simultaneous restriction of responsible vehicle and driver sets based on target routes. Especially in short-distance transportation scenarios, many orders, although belonging to the same business area, often suffer from inconsistent address descriptions due to different template sources and location text. This often results in the system being able to identify broad address categories but struggling to reliably pinpoint the intended responsible route and consequently limit the corresponding responsible vehicle and driver resources, leading to significant manual intervention in subsequent scheduling.

[0004] Another relevant prior art is Chinese patent CN104484790A, which discloses an address matching method and apparatus for logistics operations. This document proposes that the address text is first read in, then segmented according to place names in an address database to obtain administrative division information and point-of-interest (POI) information. Subsequently, the administrative division information is matched with the administrative divisions in the address database to determine the administrative division to which the POI belongs. Then, the POI information is used for precise matching or matching degree calculation. When the matching degree is greater than a threshold, the corresponding address coordinates are obtained, and it is further determined whether the address coordinates are within the delivery range, thus outputting the corresponding delivery point name. This solution reflects that current logistics information processing technology already places considerable emphasis on the standardization of address text, address matching efficiency, and address matching accuracy. It also illustrates that converting textual descriptions of addresses into calculable standard addresses or spatial locations is an important prerequisite for automating subsequent logistics processing. However, the core objective of this patent is to solve the matching problem between logistics address text and delivery address coordinates, focusing on address location and delivery point determination based on administrative divisions, POIs, and matching degree thresholds. Based on its publicly available information, this patent primarily addresses determining whether an order address falls within a specific delivery area or mapping textual addresses to coordinate delivery points. It doesn't address the more complex aspects of short-distance transportation scheduling, such as route selection, vehicle screening, driver selection, carpooling, return trip reuse, failure tracking, and manual route rerouting feedback – a complete closed loop. Specifically, when a company has a large number of short-distance orders with industry-standard wording, departmental abbreviations, and template shortenings, while relying solely on generic address segmentation and point-of-interest matching can improve text recognition to some extent, it still struggles to directly resolve the route normalization instability caused by differences in order templates. It also fails to address how to only partially update relevant dictionary fragments and route resource indexes after location text correction. In other words, this document focuses more on whether an address can be identified, rather than how the address recognition result can be tightly coupled with the responsible route, responsible vehicle, responsible driver, and subsequent scheduling rules.

[0005] In summary, existing technologies have explored related issues from at least two directions: First, the solution represented by CN105512747B emphasizes the automation of logistics scheduling through route management, address management, carpooling rule management, and capacity management; second, the solution represented by CN104484790A emphasizes improving the efficiency and accuracy of address recognition through address text standardization and matching degree judgment. While the aforementioned existing technologies have made some progress in scheduling organization and address resolution, they still have the following shortcomings for the more specific application scenario of enterprise short-distance transportation: First, there is a lack of a location semantic normalization mechanism for multi-source order templates, making it difficult to stably handle multiple text expressions for the same location; second, there is a lack of a constraint mechanism to further link route recognition results to the scope of responsible vehicles and drivers, resulting in an overly broad scheduling resource pool and frequent manual intervention; third, there is a lack of a closed-loop mechanism that forms local feedback based on manual route changes and only locally updates the relevant dictionaries and route indexes, resulting in insufficient system adaptability; and fourth, there is a lack of a dedicated scheduling framework for short-distance transportation that unifies text matching, route recognition, resource pool limitation, carpooling and return trip reuse, and failure logging.

[0006] Therefore, to address the aforementioned shortcomings of existing technologies in enterprise short-distance transportation scheduling scenarios, there is an urgent need to propose a new intelligent scheduling technology solution for short-distance transportation orders. This solution should be able to perform more targeted semantic normalization on non-standard origin and destination texts in multi-template orders, further limit the scope of responsible vehicles and drivers after route determination, and form a traceable, feedback-enabled, and locally updatable closed-loop processing mechanism after manual correction. This would improve route recognition accuracy, reduce the probability of incorrect vehicle dispatch, enhance the utilization rate of short-distance transportation resources, and strengthen the stability and maintainability of the system in actual business scenarios. Summary of the Invention

[0007] This invention aims to provide a method, system, and storage medium for intelligent scheduling of short-distance transportation orders based on template fingerprint grouping feedback. It addresses existing problems in short-distance transportation scheduling, such as inconsistent origin and destination text expressions under multi-source order templates, insufficient accuracy of automatic route identification, overly broad selection range of vehicle and driver resources, underutilization of carpooling and return trips, difficulty in generating effective feedback from manual route changes, and insufficient traceability and auditing capabilities. This invention achieves accurate and normalized identification of transportation orders, refined resource allocation under the constraint of responsible routes, and closed-loop optimization of scheduling rules, thereby improving the automation level, resource utilization efficiency, and operational stability of short-distance transportation scheduling.

[0008] Firstly, in order to achieve the above-mentioned objectives, the present invention adopts the following technical solution:

[0009] A method for intelligent scheduling of short-distance transportation orders based on template fingerprint grouping feedback, comprising the following steps:

[0010] S1. Establish a route responsibility semantic index library, including: route base table, route responsibility association table, material compatibility rule table, return route mapping table, location synonym sub-dictionary grouped by template fingerprint and material category, and responsibility inverted index associated with the route base table;

[0011] S2. Import the transportation order file and extract the order number, material category, order type, original origin text, original destination text, arrival date and required weight. Generate template fingerprints and call the corresponding location thesaurus to normalize the original origin text and original destination text to obtain the standard origin candidate set and standard destination candidate set. Then generate the candidate route set based on the responsibility inverted index.

[0012] S3. When the candidate route set meets the preset matching acceptance conditions, the target route is determined, and a restricted candidate resource pool is generated by the set of responsible vehicles, the set of responsible drivers and the remaining load constraints of the vehicles corresponding to the target route; otherwise, it is written into the queue to be corrected.

[0013] S4. Perform hierarchical scheduling only within the restricted candidate resource pool in the order of direct dispatch of whole vehicles, compatible carpooling, return trip reuse, and overload splitting, and output the scheduling result or the reason for scheduling failure.

[0014] S5. When a user manually reroutes an order in the order to be corrected or a failed order, a correction sample is formed. When the correction sample meets the preset dictionary update trigger condition, only the corresponding location synonym sub-dictionary is updated and the relevant responsibility inverted index local segment is reconstructed.

[0015] Preferably, in step S1, the route base table records the standard starting point, standard ending point, and route identifier; the route responsibility association table records the set of responsible vehicles and the set of responsible drivers corresponding to each route identifier; the index key of the responsibility inverted index includes at least the standard starting point, standard ending point, and material category; and the index value includes at least the route identifier, the set of responsible vehicles, the set of responsible drivers, the material compatibility rule, and the return route identifier.

[0016] The responsible vehicle set and the responsible driver set are stored using a responsible vehicle bitmap and a responsible driver bitmap, respectively; each index record of the responsibility inverted index also includes an available load limit field, a return route identifier field, and a material compatibility code field; wherein, the responsible vehicle bitmap is a binary bitmap representing the relationship between the responsible vehicle members for a certain route; the responsible driver bitmap is a binary bitmap representing the relationship between the responsible driver members for a certain route;

[0017] The template fingerprint is generated by combining the field title sequence, worksheet name, source unit identifier, and order type of the transportation order file. It is used as the grouping key of the location thesaurus to distinguish the differences in location abbreviations, acronyms, or customary spellings under different order templates.

[0018] As a preferred option, in step S2, the target route is determined when the candidate route set meets the preset matching acceptance conditions; otherwise, the order is written to the queue to be corrected and transportation resources are not allocated temporarily.

[0019] And / or, in step S2, the route matching confidence score is calculated for each candidate route, using the following formula:

[0020] ;

[0021] in, Calculate the route matching confidence score for candidate routes; The score is the match score between the original starting text and the standard starting text; The score is the score for matching the original endpoint text with the standard endpoint; A score is awarded for the consistency between the material category and the route. , and For the corresponding weights, and satisfying ;

[0022] When the route matching confidence of the highest candidate route is The route matching confidence of the second-highest candidate route is At that time, only Not less than the matching confidence threshold and Not less than the confidence difference threshold In the case of [condition], the candidate route set is determined to meet the preset matching acceptance conditions.

[0023] Preferably, in step S3, the restricted candidate resource pool is generated through the following intersection operation:

[0024] ;

[0025] ;

[0026] in, For candidate vehicle bitmaps; Location map of the responsible vehicles for the target route; This is a bitmap showing currently available vehicles. A bitmap of candidate drivers; A map showing the locations of the responsible drivers for the target route; This is a bitmap of currently available drivers; To match the candidate vehicle bitmap The driver's location map that can be bound to the vehicle; only the aforementioned and The corresponding vehicles and drivers are written into the restricted candidate resource pool.

[0027] Preferably, in step S4, the overload is split into multiple sub-orders that inherit the standard start point, standard end point, material category and arrival date of the original order, where the required weight exceeds the available load capacity of a single vehicle.

[0028] And / or, in step S4, direct dispatch of the whole vehicle is executed when the required weight of a single order is not greater than the available load capacity of a single vehicle and there are no compatible orders that can be combined.

[0029] Compatible carpooling is implemented when multiple orders have the same destination route, the material compatibility codes are consistent or mutually compatible, the total demand weight after merging is not greater than the available load capacity of a single vehicle, and the difference in arrival dates is not greater than the preset carpooling date window.

[0030] Return trip reuse is performed when the current return route identifier of the candidate vehicle and the return route identifier of the target route are mutually mapped, and the expected return time of the candidate vehicle falls within the preset return time window;

[0031] Overload splitting is performed when the order demand weight exceeds the single-vehicle available load capacity of any vehicle in the restricted candidate resource pool.

[0032] Preferably, steps S4 and S5 further include: writing automatic scheduling operations, manual route rerouting operations, and scheduling failure operations into historical records; the historical records include at least the record type, order number, original origin text, original destination text, target route identifier, vehicle identifier, driver identifier, reason for failure, operator, operation time, and operation result; the reasons for scheduling failure include at least route not identified, insufficient responsible vehicles, insufficient responsible drivers, incompatible materials, overload, and failure to meet return conditions.

[0033] Preferably, in step S5, the template fingerprint, original starting point text, original ending point text, material category, corrected route identifier, operation time, and operation result are recorded to form a correction sample; when the correction sample corresponding to the same original starting point text, original ending point text, and material category under the same template fingerprint meets the preset dictionary update trigger condition, only the location synonym sub-dictionary corresponding to the template fingerprint is updated, and the local fragment of the responsibility inverted index associated with the updated term is reconstructed for automatic route matching and resource pool limitation of subsequent orders;

[0034] And / or, in step S5, the preset dictionary update trigger condition is determined by a corrected consistency coefficient, the formula for which the corrected consistency coefficient is calculated is:

[0035] ;

[0036] in, To correct the consistency coefficient; In the statistics window Within, the number of corrected samples that have the same template fingerprint, the same original starting point text, the same original ending point text, and the same material category, and have been manually changed to the same route identifier; In the statistical window Within this range, the total number of corrected samples with the same template fingerprint, the same original starting text, the same original ending text, and the same material category; For statistical windows;

[0037] Only when Not less than the sample size threshold and Not less than the consistency threshold Only then is the location thesaurus updated and the corresponding responsibility inverted index segment reconstructed.

[0038] Secondly, the present invention also provides an intelligent scheduling system for short-distance transportation orders based on template fingerprint grouping feedback. This system is used to implement the method described above, including:

[0039] The route responsibility semantic index module is used to establish and maintain the route base table, route responsibility association table, material compatibility rule table, return route mapping table, location thesaurus, and responsibility inverted index.

[0040] The order import and template fingerprint generation module is used to import shipping order files, extract order fields, and generate template fingerprints.

[0041] The route matching module is used to normalize the original starting point text and the original ending point text based on the location thesaurus corresponding to the template fingerprint and generate a candidate route set.

[0042] The restricted resource pool generation module is used to generate a restricted candidate resource pool based on the set of responsible vehicles, the set of responsible drivers, the vehicle idle status, the driver available status, and the vehicle remaining load corresponding to the target route.

[0043] The hierarchical scheduling module is used to perform scheduling only within the restricted candidate resource pool in the order of direct dispatch of whole vehicles, compatible carpooling, return trip reuse, and overload splitting.

[0044] The feedback update module is used to record the correction samples formed by manual route changes, and when the preset dictionary update trigger conditions are met, it only updates the corresponding location thesynonym sub-dictionary and reconstructs the local fragment of the responsibility inverted index.

[0045] The history tracking module is used to record information about the entire process of automatic scheduling, manual route rerouting, and scheduling failures.

[0046] Thirdly, the present invention also provides an electronic device, including a processor, a memory, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, it implements the method described above.

[0047] Fourthly, the present invention also provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the method described thereon.

[0048] This invention constructs a sub-dictionary of location synonyms grouped by template fingerprints and material categories, and combines this with a responsibility-based inverted index to normalize and match the original origin and destination texts in orders with the target route. This effectively reduces the impact of different order templates, different filling habits, and differences in location abbreviations on the accuracy of route identification, significantly improving the stability of automatic matching under non-standard text conditions. Furthermore, after determining the target route, this invention further restricts scheduling resources by using a set of responsible vehicles and a set of responsible drivers, and generates a restricted candidate resource pool by combining vehicle idle status, driver availability status, and vehicle remaining load capacity. This avoids coarse screening of the entire transportation capacity, narrows the scheduling search range, and improves the targeting and execution efficiency of vehicle and driver allocation. In addition, this invention only performs hierarchical scheduling within the restricted candidate resource pool, in the order of direct dispatch of whole vehicles, compatible carpooling, return trip reuse, and overload splitting. This invention not only improves vehicle loading rate and return vehicle utilization rate, reduces empty running rate and scheduling failure rate, but also provides a more reasonable splitting and failure reason output mechanism when order weight exceeds limits or resources are insufficient, enhancing the interpretability of system scheduling results. Furthermore, this invention samples and records the results of manual route changes, and only updates the local fragments of the location synonym sub-dictionary and related responsibility inverted index under the corresponding template fingerprint when preset update conditions are met. This can gradually improve the system's ability to adapt to actual business writing habits, and avoid the risk of mismatch diffusion caused by frequent updates of the global dictionary. Therefore, it realizes a closed-loop processing of short-distance transportation orders from semantic recognition, route determination, resource constraints, scheduling execution to feedback optimization, and has beneficial technical effects such as improving the level of scheduling automation, reducing the intensity of manual intervention, enhancing operational stability, improving resource utilization efficiency, and strengthening the ability to leave traces throughout the entire process. Attached Figure Description

[0049] Figure 1 This is a schematic diagram of the overall structure of the intelligent scheduling system for short-distance transportation orders based on template fingerprint grouping feedback according to the present invention.

[0050] Figure 2This is a schematic diagram illustrating the composition and data association relationships of the route responsibility semantic index library of this invention.

[0051] Figure 3 This is a schematic diagram of the order import, template fingerprint generation, and field standardization process of the present invention.

[0052] Figure 4 This is a schematic diagram illustrating the process of starting point and ending point normalization and candidate route generation based on a location thesaurus of the present invention.

[0053] Figure 5 This is a schematic diagram of the processing flow of the present invention for generating a restricted candidate resource pool based on the responsible vehicle bitmap and the responsible driver bitmap.

[0054] Figure 6 This is a schematic diagram illustrating the hierarchical scheduling process of the present invention, which executes the hierarchical scheduling in the order of direct dispatch of whole vehicles, compatible carpooling, reuse of return trip vehicles, and splitting of overloaded vehicles.

[0055] Figure 7 This is a schematic diagram of the closed-loop feedback process of the present invention, which is based on manual route rerouting to form corrected samples and triggers local updates of the location synonym sub-dictionary and local reconstruction of the responsibility inverted index.

[0056] Figure 8 This is a schematic diagram of the scheduling result output, scheduling failure reason recording, and historical trace tracking process of the present invention.

[0057] Figure 9 A comparison chart showing the accuracy of automatic route identification versus the rate of manual route modification under different schemes.

[0058] Figure 10 This is a comparison chart of average scheduling time under global search and limited candidate resource pool methods.

[0059] Figure 11 This is a comparison chart of the average loading rate and empty-running rate under Comparative Example 2 and the scheme of the present invention.

[0060] Figure 12 The graph shows the improvement in the accuracy of automatic route recognition within 4 weeks after the local feedback update. Detailed Implementation

[0061] The technical solutions in the embodiments of the present invention will be clearly and completely described below. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the protection scope of the present invention.

[0062] This invention provides a method, system, and storage medium for intelligent scheduling of short-distance transportation orders based on template fingerprint grouping feedback. It is applicable to scenarios such as raw material distribution within enterprise parks, transshipment transportation, inter-process turnover transportation, and short-distance finished product transfer. This solution is not a simple aggregation of general logistics scheduling rules, but rather addresses the core issues of inconsistent location text due to differences in order templates, the disconnect between location identification and responsibility capacity constraints, and the difficulty in accumulating manual route-changing experience. It constructs a closed-loop technical route consisting of template fingerprint grouping, sub-dictionary normalization, responsibility-based inverted index retrieval, generation of vehicle / driver dual-constraint resource pools, hierarchical scheduling, and local feedback updates. This closed-loop technical route corresponds to the business modules described in the manual, such as order import, route management, vehicle management, driver management, intelligent scheduling, and historical record keeping, but is further refined through patented technology, enabling those skilled in the art to implement it without creative effort.

[0063] I. Terminology Explanation

[0064] Template fingerprint: A unique or quasi-unique identifier generated based on template features such as field title sequence, worksheet name, source unit identifier, and order type in a shipping order file. It is used to distinguish order templates from different departments, sources, and styles. The template fingerprint does not directly represent the specific order content, but rather the structural characteristics of the order's source template.

[0065] Location thesaurus: This refers to a set of local location semantic mappings formed by further grouping according to template fingerprints and material categories within the overall dictionary system. This sub-dictionary is used to map common location abbreviations, aliases, historical names, or acronyms under the same template to standard starting points or standard ending points.

[0066] Responsibility Inverted Index: This refers to an index structure built based on standard starting point, standard ending point, and material category. Its index value includes at least route identifier, set of responsible vehicles, set of responsible drivers, material compatibility rules, and return route identifier. It is used to quickly obtain candidate routes and their limited scheduling resources after location normalization.

[0067] Responsible vehicle bitmap: refers to a binary structure formed by bit-level encoding of the set of vehicles responsible for a certain route, where each bit corresponds to the registration identifier of a vehicle, used to quickly determine whether a vehicle belongs to the scope of responsible vehicles for that route.

[0068] Responsible Driver Bitmap: This refers to a binary structure formed by bit-level encoding of the set of drivers responsible for a certain route. Each bit corresponds to a driver's registration identifier, which is used to quickly determine whether a driver belongs to the scope of responsible drivers for that route.

[0069] Limited candidate resource pool: This refers to the set of candidate vehicles and drivers selected after the target route is determined, based on the responsible vehicle bitmap, vehicle idle status bitmap, responsible driver bitmap, driver available status bitmap, and vehicle-driver binding relationships. This resource pool is not a global resource set, but a local resource set strongly bound to the target route.

[0070] Route matching confidence score: For a candidate route, the comprehensive score obtained by fusing the starting point matching score, the ending point matching score, and the material category consistency score according to preset weights is used to determine whether the automatic route identification result is reliable.

[0071] Return route mapping: refers to the mapping relationship between forward routes and reverse return routes established in the route management module. It is used to determine whether a vehicle can undertake a new short-distance transportation task under return or near-return conditions after completing a certain route transportation.

[0072] Correction Sample: refers to the structured record formed after a failed order is manually rerouted, confirmed, or corrected. It includes at least the following fields: template fingerprint, original starting point text, original ending point text, material category, corrected route identifier, operation time, and operation result.

[0073] Correction Consistency Coefficient: Within a predetermined statistical window, among the corrected samples with the same template fingerprint, the same original starting text, the same original ending text, and the same material category, the proportion of samples that were ultimately manually rerouted to the same route identifier is used to determine whether a location synonym sub-dictionary update should be triggered.

[0074] II. System Structure

[0075] like Figure 1 As shown, the system architecture of the present invention may include a front-end interaction layer, a business processing layer, and a data storage layer.

[0076] The front-end interaction layer may include a login interface, homepage dashboard, order management page, intelligent dispatch page, vehicle management page, route management page, driver management page, historical record page, and system management page. This layer is mainly used to receive order files uploaded by users, display parsing results, display dispatch results, receive manual route modification operations, and output historical record information. Its interface organization is consistent with the order management, intelligent dispatch, vehicle management, route management, driver management, and historical record pages shown in the manual.

[0077] The business processing layer may include an order import and parsing module, a template fingerprint generation module, a location normalization module, a candidate route generation module, a restricted candidate resource pool generation module, a hierarchical scheduling module, a manual feedback update module, and a history tracking module. Among these:

[0078] The order import and parsing module is used to read Excel format shipping order files, standardize and identify fields, and generate structured order records;

[0079] The template fingerprint generation module is used to generate template fingerprints based on the structural features of the order template;

[0080] The location normalization module is used to call the location synonym sub-dictionary corresponding to the template fingerprint and material category to perform semantic normalization on the original starting text and the original ending text.

[0081] The candidate route generation module is used to form a candidate route set based on the responsibility inverted index and normalization results, and to calculate the route matching confidence.

[0082] The restricted candidate resource pool generation module is used to generate a local candidate resource set based on the set of responsible vehicles, the set of responsible drivers, and the real-time availability status of the target route.

[0083] The hierarchical scheduling module is used to perform direct dispatch of whole vehicles, compatible carpooling, return trip vehicle reuse, and overload splitting;

[0084] The manual feedback update module is used to generate correction samples based on manual route rerouting operations, and update the location thesaurus and local responsibility inverted index when conditions are met;

[0085] The history tracking module is used to record various operations, results, and reasons for failure.

[0086] The data storage layer may include an order data table, a vehicle information table, a driver information table, a route basic table, a route responsibility association table, a material compatibility rule table, a return route mapping table, a location thesaurus table, a responsibility inverted index table, a correction sample table, and a history record table. Through the organization of these data tables, this invention can not only manage existing orders, vehicles, routes, drivers, and historical records in the document, but also support the newly added template fingerprint grouping feedback and local index update mechanisms.

[0087] III. Overall Technical Route of the Method of the Invention

[0088] like Figures 3 to 8As shown, the method of the present invention can be executed according to steps S1 to S5. Its overall technical approach is as follows: First, a route responsibility semantic index library is established; then, transportation order files are imported and template fingerprints are generated, and the origin and destination normalization and candidate route generation are completed using a location synonym sub-dictionary; subsequently, responsible vehicles and drivers are synchronously constrained according to the target route to form a restricted candidate resource pool; then, hierarchical scheduling such as whole-vehicle, carpooling, return trip, and splitting is performed only within the restricted candidate resource pool; finally, manual route modification behavior is sampled and recorded, and when triggering conditions are met, only the sub-dictionary and local responsibility inverted index under the corresponding template are updated, and the entire process is written into the historical record.

[0089] This technical approach differs from general global dictionary matching, global vehicle search, and experience-based dispatching. This invention uses template fingerprints as a semantic isolation dimension, local sub-dictionaries as a location normalization carrier, responsibility inverted index as a link between routes and resources, vehicle / driver bi-bitmap intersection as a resource pool shrinkage mechanism, and local feedback update as an adaptive evolution mechanism. Therefore, it has better adaptability in non-standard text, multiple template sources, and short-distance responsibility transportation scenarios.

[0090] (I) Step S1: Establish a route responsibility semantic index library

[0091] Step S1 is one of the key innovative steps of this invention. Its function is not merely to establish a standard route configuration table, but rather, during system initialization or administrator maintenance, to construct a data organization system that simultaneously includes route semantics, responsible resources, material compatibility relationships, and return route mapping relationships. This lays the foundation for subsequent automatic route matching and resource constraints. The technical contributions of this step are mainly twofold: firstly, routes are no longer simply records of origin, destination, and distance, but are expanded into multi-dimensional scheduling objects encompassing standard origin, standard destination, material category, set of responsible vehicles, set of responsible drivers, return route, and compatibility rules; secondly, these data are not scattered but interconnected through a responsibility-based inverted index, allowing direct retrieval of the target route and its corresponding resource range after order text normalization.

[0092] Specifically, such as Figure 2 As shown, the route responsibility semantic index library may include at least the following data units:

[0093] First, the route baseline table. This table stores standardized route entities. Each record includes at least a route identifier, standard start-point code, standard end-point code, route name, route status, mileage or distance information, default estimated duration, and the business area it belongs to. When implemented within a corporate campus, the standard start-point and standard end-point should preferably use a unified internal coding system, such as warehouse area code, workshop code, storage code, or gate code, to avoid ambiguity caused by differences in Chinese names. The route status in the route baseline table can include enabled, disabled, and under maintenance statuses. Disabled routes do not participate in automatic scheduling.

[0094] Second, the route responsibility association table. This table associates each route with a set of responsible vehicles and a set of responsible drivers. The set of responsible vehicles can consist of one or more vehicle identifiers, and the set of responsible drivers can consist of one or more driver identifiers. For some routes, the set of responsible vehicles can be further subdivided into a primary responsible vehicle set and a backup responsible vehicle set; the set of responsible drivers can also be subdivided into a regular responsible driver set and a reserve driver set. For ease of rapid calculation, this embodiment preferably encodes the set of responsible vehicles and the set of responsible drivers as a responsible vehicle bitmap and a responsible driver bitmap, respectively. The bitmap length can be equal to the total number of currently active vehicles and the total number of currently active drivers, respectively. Each bit corresponds to the internal sequence number of a vehicle or driver. If a vehicle or driver falls within the responsibility scope of the route, the corresponding bit is recorded as 1; otherwise, it is recorded as 0.

[0095] Third, the Material Compatibility Rule Table. This table defines whether different material categories can be transported together in a shared vehicle. For example, material categories can be divided into carbon black, rubber materials, skeleton materials, auxiliary materials, empty pallets, etc., and a compatibility code can be assigned to each category. If two orders have the same compatibility code, or are marked as compatible in the compatibility matrix, they are allowed to enter the shared vehicle candidate group; if they are marked as incompatible, they are prohibited from being transported together in the same vehicle. This rule corresponds to the function in the Material Type Overview section of the manual that supports shared vehicle rules for different material types.

[0096] Fourth, the return route mapping table. This table records the return route identifier or approximate return route identifier corresponding to a forward route. For example, if route R001 represents the raw material warehouse → rubber mixing workshop, its return route can be mapped to rubber mixing workshop → raw material warehouse. For situations where there are circular roads or different entrances / exits within the park, the rubber mixing workshop → raw material transfer area can also be defined as an approximate return route. When the system performs return vehicle reuse, it can quickly determine through this mapping table whether a vehicle that has completed the previous task can undertake subsequent tasks in the return or near-return direction.

[0097] Fifth, the location synonym sub-dictionary table. This table is an important component of this invention. Unlike traditional global synonym dictionaries, this invention does not uniformly store location aliases under all departments, templates, and material categories. Instead, it uses template fingerprints and material categories as grouping keys to form multiple local sub-dictionaries. Each sub-dictionary entry includes at least the original location entry, standard location code, standard location name, entry source, creation time, last update time, validity status, and update count. The entry source can include three categories: system preset, manual maintenance, and manual feedback verification. The advantage of using the sub-dictionary method is that the same abbreviation under different templates may not correspond to the same location. For example, "South Warehouse" in Department A's template may correspond to "South Raw Material Warehouse," while "South Warehouse" in Department B's template may correspond to "South Finished Product Temporary Storage Warehouse." If a globally unified mapping is used, it is easy to cause cross-template contamination. The sub-dictionary grouped by template fingerprints can effectively avoid the propagation of such errors.

[0098] Sixth, the Responsibility Inverted Index Table. This table uses the standard origin, standard destination, and material category as index keys, and route identifiers, responsible vehicle location maps, responsible driver location maps, material compatibility codes, and return route identifiers as index values. The inverted index means that instead of traversing each route one by one to check for matching with the order, it directly matches all possible route records based on the normalized origin, destination, and material category. This way, once the order's origin and destination are normalized to standard locations, candidate routes can be obtained in constant or near-constant time without requiring a full comparison of all routes.

[0099] In actual deployment, step S1 can be completed by the administrator entering or importing basic data in the route management, vehicle management, and driver management pages. Specifically, the basic route information, distance matrix, and route material type correspondence can be entered first in the route management module. Then, information such as the vehicle's rated load capacity, fleet affiliation, permanent location, and current status can be entered in the vehicle management module. In the driver management module, information such as driver qualifications, shifts, contact information, and executable route set can be entered. Finally, the backend service writes the configuration relationship between routes and vehicles, and routes and drivers, into the route responsibility association table, and simultaneously generates a responsible vehicle location map and a responsible driver location map. This process corresponds to the freely configurable transportation parameters and functional modules such as route management, vehicle management, and driver management in the manual.

[0100] Preferably, during the initial database construction, an index fragment can be generated for each route. Specifically, the standard start-point code, standard end-point code, and applicable material category of a route are read; the responsible vehicle and driver bitmaps are read from the route responsibility association table; the compatibility code for the material category is read from the material compatibility rule table; and the return route identifier is read from the return route mapping table. Then, these fields are assembled into a responsibility inverted index record and written to the index table. If a route allows the transportation of multiple material categories, multiple index records can be generated separately. If a route has both a primary responsibility set and a backup responsibility set, index value fields with different priorities can be generated so that primary responsibility resources are retrieved first, and backup responsibility resources are retrieved only when necessary during subsequent scheduling.

[0101] To improve the efficiency of subsequent partial updates, this implementation also preferably records the index fragment version number and index source field in the responsibility inverted index table. The index fragment version number is used to identify whether the index was generated during the initial database creation by the system or rebuilt due to updates triggered by manual feedback; the index source field is used to identify the version of its corresponding template fingerprint sub-dictionary. In this way, when the location synonym sub-dictionary under a certain template fingerprint changes subsequently, it is not necessary to rebuild the global index. Instead, it is only necessary to delete the old version partial fragment and write the new version partial fragment according to the affected standard locations and material categories.

[0102] The route responsibility semantic index library established in the above manner can not only support ordinary route configuration and scheduling, but also provide a unified data foundation for subsequent template group semantic recognition, local feedback updates and restricted resource pool generation. This is one of the key differences between this invention and traditional transportation scheduling systems.

[0103] (II) Step S2: Order import, template fingerprint generation and candidate route generation

[0104] Step S2 is another highly significant step in this invention. Its key lies not in simply reading the Excel file, but in extracting the structural features of order templates from different sources, with different structures, and different naming conventions. Then, it uses the template fingerprint to call a local location thesaurus to complete semantic normalization, and finally generates candidate routes through a responsibility-based inverted index. This method isolates template differences from location identification errors, avoiding misidentification caused by global matching.

[0105] like Figure 3 and Figure 4 As shown, step S2 can be further divided into the following sub-steps.

[0106] 1. Order file import and field recognition

[0107] After a user clicks the "Import Order" button on the order management page, the system can receive Excel format shipping order files, supporting drag-and-drop upload or file selection upload. After reading the workbook, the system scans each worksheet separately, identifying the header area, data start row, field header row, and data end row. During field identification, a set of standard field mapping rules can be set. For example, the origin (shipping location, shipping warehouse, loading point) can be uniformly mapped to the original origin field; the destination (receiving location, unloading point, destination warehouse) can be uniformly mapped to the original destination field; material category, material type, cargo type can be uniformly mapped to the material category field; and tonnage, quantity, and required weight can be uniformly mapped to the required weight field.

[0108] To ensure recognition stability, preprocessing such as full-width / half-width conversion, case normalization, bracket denoising, and special character cleaning can be performed on field titles. If a field title cannot be uniquely mapped to a standard field, the corresponding order in that worksheet is marked as a template to be verified. If all standard fields are present, the order content is parsed into structured records, written to a temporary order cache table, and the parsing results are displayed on the front end. This corresponds to the function in the manual where clicking the "Start Parsing All Files" button allows the system to extract and process keywords from uploaded and imported demand orders, and display an overview of the parsed demand data.

[0109] 2. Template fingerprint generation

[0110] In this invention, the template fingerprint is preferably generated by combining the following template features: a sequence of field titles, a worksheet name, a source unit identifier, an order type identifier, and optional file naming features. Specifically, the field titles are first standardized from left to right to form a title sequence string; then the worksheet name is read; if the file contains a source unit field, the source unit is extracted; if the order contains an order type field, the order type is extracted; finally, these contents are concatenated in a fixed order to form a template feature string, and a template fingerprint is generated using a hash function, a verification function, or other deterministic encoding algorithms.

[0111] For example, a worksheet uploaded by a purchasing department might be named "Raw Material Daily Plan," with field headings including Required Date, Shipping Location, Receiving Location, Material, Weight, and Remarks. The source unit is the Purchasing and Dispatch Office, and the order type is "Raw Material Replenishment." The system can then concatenate these contents to generate template fingerprint T001. A file uploaded by another warehousing department might have field headings including Date, Origin Point, Destination Point, Cargo Type, Tonnage, and Workshop. Even if its business meaning is similar to the former, it will generate a different template fingerprint, T002. During subsequent location synonym matching, the sub-dictionaries called by T001 and T002 are not interchangeable, thus avoiding term contamination caused by differences in writing styles between different departments.

[0112] 3. Original Location Text Preprocessing and Sub-Dictionary Calling

[0113] After generating the template fingerprint, the system reads the material category of the order and uses the template fingerprint + material category as the union key to call the corresponding location thesaurus. Before calling, the system preprocesses the original starting and ending texts. Preprocessing may include: removing leading and trailing spaces, unifying Chinese and Arabic numerals, removing meaningless stop words, merging repeated punctuation, and unifying workshops (e.g., workshop 1# workshop) into a comparable intermediate form. Then, the intermediate form is fed into a location segmenter or rule segmenter to obtain one or more location tokens.

[0114] If a term in the sub-dictionary perfectly matches a location term, the corresponding standard location code is output directly. If no perfectly matching term exists, fuzzy matching, prefix matching, or historical manual feedback mapping matching can be performed. For the same original location text, multiple standard location candidates may be obtained. In this case, the system will form a standard starting point candidate set and a standard ending point candidate set, rather than immediately determining a unique location. This mechanism leaves room for subsequent route-level discrimination based on a responsibility-based inverted index.

[0115] It should be noted that, unlike a general global dictionary, the sub-dictionaries of this invention are grouped according to template fingerprints and material categories. For example, when the template fingerprint is T001 and the material category is carbon black, the CB library's carbon black warehouse can be mapped to the standard starting code S_CB_01; however, when the template fingerprint is T002 and the material category is finished product, the term CB library may not exist at all, or may be mapped to other library locations. It is this template-level and material-level local mapping that enables this invention to achieve higher semantic parsing accuracy in the high-frequency but non-standardized scenario of short-distance transportation.

[0116] 4. Candidate route set generation and route matching confidence calculation

[0117] For each standard starting point in the candidate starting point set and each standard ending point in the candidate ending point set, the system can create an index key with the current material category and query it using the inverted responsibility index. If a corresponding record exists in the index table, the route identifier in that record is extracted to form a candidate route set. If different starting-ending point combinations hit the same route, the hit count and source of that route are merged and recorded.

[0118] Subsequently, the system calculates the route matching confidence score for each route in the candidate route set. Preferably, the following formula can be used:

[0119] ;

[0120] in, Calculate the route matching confidence score for candidate routes; The score is calculated based on the match between the original starting text and the standard starting point of the route. The score is assigned to the match between the original endpoint text and the standard endpoint of the route. A score is given for the consistency between the material category and the material categories allowed for this route; , and For the corresponding weights, and satisfying .

[0121] In a preferred embodiment and The score can be determined as follows: if the original text is exactly the same as the standard location name, the score is 1; if it is hit by a specific mapping term in the sub-dictionary, the score is 0.95; if it is hit by a fuzzy rule or word combination, the score is 0.80 to 0.90; if there is only a weak association, the score can be set to 0.60 to 0.75. The value can be 1 when the material categories are completely identical; 0.80 when they belong to the same superior material family but are not completely identical; 0.60 when there are restrictions but transportation is still allowed; and 0 when not allowed.

[0122] The highest candidate route matching confidence score after sorting and the second highest candidate route matching confidence The system preferentially determines whether to automatically accept based on the following rules:

[0123] when and At that time, it is considered that the candidate route set meets the preset matching acceptance conditions.

[0124] in, To match the confidence threshold, it is preferably set to 0.80 to 0.90; The confidence difference threshold is preferably set to 0.05 to 0.20. If the confidence level of the highest route match does not reach the threshold, or the difference between the highest and second-highest routes is too small, it is considered that there is uncertainty in the automatic route identification. The order will not be entered into the automatic dispatching system, but will be written into the correction queue, waiting for manual review or confirmation.

[0125] (III) Step S3: Generate a restricted candidate resource pool

[0126] The purpose of step S3 is to generate a smaller, more precise pool of restricted candidate resources by leveraging the dual intersection of responsibility sets and real-time status, rather than blindly filtering among all vehicles and drivers after determining the target route. Although this step employs the conventional technique of bitmap intersection, its combination with the route responsibility index is a crucial intermediate step in achieving efficient and stable scheduling in this invention.

[0127] First, the system reads the responsible vehicle bitmap corresponding to the target route from the inverted responsibility index records. and driver's position diagram Subsequently, a bitmap of currently available vehicles is generated from the vehicle status table and the driver status table, respectively. and currently available driver bitmap Vehicle status can include idle, assigned, en route, under maintenance, and out of service, with only the bit corresponding to an idle vehicle being set to 1; driver status can include available, dispatched, resting, on leave, and out of service, with only the bit corresponding to an available driver being set to 1.

[0128] To ensure the executable relationship between vehicles and drivers, the system can also read the vehicle-driver binding table or qualification matching table to form a driver binding bitmap. .in, This represents the set of drivers who can be legally matched with the current set of candidate vehicles. For example, some large vehicles require drivers with an A2 driving license, and some forklift routes are only permitted to be operated by drivers from specific shifts; in this case, drivers who do not meet the qualifications... The corresponding position is 0.

[0129] The restricted candidate resource pool can be generated according to the following bitmap operation:

[0130] ;

[0131] ;

[0132] in, For candidate vehicle bitmaps; Location map of the responsible vehicles for the target route; This is a bitmap showing currently available vehicles. A bitmap of candidate drivers; A map showing the locations of the responsible drivers for the target route; This is a bitmap of currently available drivers; To match the candidate vehicle bitmap Driver location map that can be bound to a vehicle.

[0133] get and Next, the system reads the vehicle and driver entity information positioned in the bitmap one by one to form a candidate vehicle list and a candidate driver list, and further filters them according to conditions such as the vehicle's remaining load, the vehicle's current location, the estimated empty driving distance, and the driver's remaining shift time. The vehicle's remaining load can be obtained by subtracting the load it has already undertaken but not yet completed from its rated load. If a vehicle has a rated load of 10 tons, and there are currently 2 tons of unloaded load, then its remaining load is 8 tons.

[0134] Preferably, the system can organize vehicle entity records into the following fields: vehicle identifier, license plate number, vehicle type, rated load capacity, current status, current location, allocated load capacity, fleet to which it belongs, and most recent available time; driver entity records can be organized into: driver identifier, name, qualification level, current status, current shift, bound vehicle set, team to which it belongs, and most recent available time. In the restricted candidate resource pool, each resource pair must include at least the vehicle identifier, driver identifier, available load capacity per vehicle, estimated arrival time at the origin, and resource priority.

[0135] In one embodiment, if the location map of the responsible vehicle for the target route R015 Corresponding vehicles V01, V03, and V08, available vehicle location map For vehicles V01, V04, and V08, the intersection of the bitmaps yields... Corresponding vehicles V01 and V08. If the responsible driver's location is shown in the diagram. Corresponding to drivers D12, D14, and D16, the currently available driver bitmap Corresponding to drivers D12, D15, and D16, and driver binding bitmaps compatible with V01 and V08. Corresponding to drivers D12 and D16, the final result is... This corresponds to drivers D12 and D16. The system then combines information such as the available load capacity and current location of vehicles V01 and V08 to generate a pool of restricted candidate resources. In this way, S4 can schedule resources only within this pool, without needing to scan all vehicles and drivers globally.

[0136] This approach avoids the inefficiency of indiscriminate matching across the entire fleet after route identification. It also ensures that manually set parameters for each route's responsible vehicle fleet and corresponding driver are actually executed by the system, rather than remaining merely data displays. This aligns closely with the manual's description of users being able to set their own inherent transportation routes, responsible vehicle fleets, and corresponding driver parameters, but in this invention, it is further technically refined into a calculable and rapidly intersecting resource constraint mechanism.

[0137] (iv) S4 step: hierarchical scheduling execution

[0138] Step S4 is to complete the specific vehicle and personnel dispatch and task organization within the limited candidate resource pool. Considering that direct dispatch of whole vehicles, carpooling, utilization of return trips, and splitting of overloaded vehicles are relatively common scheduling logics in existing scheduling technologies, this step focuses on explaining their execution order in this invention and their coordination with the limited resource pool, rather than taking them as the main source of inventiveness.

[0139] like Figure 6As shown, when the system executes step S4, it can first sort the order queue according to fields such as order priority, arrival date, and import time. Orders with higher priority and earlier arrival times are processed first. Then, the following sub-steps are executed sequentially:

[0140] 1. Direct dispatch of the whole vehicle

[0141] If the required weight of an order is not greater than the available load capacity of at least one candidate vehicle in the restricted candidate resource pool, and there is no suitable compatible carpooling partner for the order within the current time window, the system will prioritize selecting a vehicle that meets the conditions and a driver that can be bound to it to form a direct dispatch plan. Candidate vehicles can be selected based on rules such as the available load capacity being closest to the required weight, the current location being closest to the origin, and the estimated empty driving time being the shortest.

[0142] 2. Compatible with carpooling

[0143] If multiple orders have the same destination route, or belong to adjacent routes that can be merged under the rules, and have the same or compatible material compatibility codes, and the difference in arrival dates between the orders does not exceed the preset carpooling date window, then the system will group these orders into the same carpooling candidate group. The carpooling date window refers to the maximum allowable difference between the promised arrival dates of several orders sharing the same vehicle; for example, it can be set to 0.5 days or 1 day. During carpooling, the system preferentially adopts a strategy of loading high-priority orders first, followed by low-priority orders, accumulating the required weight of orders until it approaches but does not exceed the available load capacity of a single vehicle. If adding the next order would cause overloading, merging will stop, and the system will continue searching for other vehicles for the remaining orders.

[0144] 3. Reuse of return trip vehicles

[0145] For target routes with return route mappings, the system checks among candidate vehicles for available vehicles that have completed their previous task and are currently on their return journey or nearing it. If the vehicle's current return route identifier and the target route are mapped in the return route mapping table, and the vehicle's estimated return time falls within a preset return time window, the system prioritizes reusing that return vehicle to perform the current task. The return time window refers to the maximum permissible deviation between the time it takes for a return vehicle to finish its previous task and be ready to load again, and the current order's allowed dispatch time. For example, it can be set to 2 hours or 4 hours. This strategy helps reduce empty-running rates and improve vehicle turnaround efficiency.

[0146] 4. Overload splitting

[0147] If the order's required weight exceeds the available payload of any vehicle in the restricted candidate resource pool, the system automatically splits the order into multiple sub-orders. The split sub-orders inherit the original order's standard origin, standard destination, material category, and arrival date, differing only in the weight field and sub-order number. Preferably, the system can prioritize splitting based on the maximum available payload of a single vehicle. For example, if the total required weight of the order is 18 tons, and the maximum available payload of a single vehicle in the restricted candidate resource pool is 8 tons, it can be split into three sub-orders: 8 tons, 8 tons, and 2 tons, and each sub-order will then be processed again using either full-vehicle direct dispatch or carpooling logic.

[0148] 5. Output the reason for failure

[0149] If the above steps fail to assign a valid vehicle or driver to an order, the system will generate a dispatch failure record. Reasons for failure may include route not identified, insufficient responsible vehicles, insufficient responsible drivers, overload, incompatible materials, unmet return conditions, incorrect route information, or the need for route updates. This corresponds to the manual's description of dispatch failure reasons, including no available vehicles, incorrect imported route information (or the need to update route information).

[0150] After scheduling is completed, the system can generate a scheduling plan display, which includes the responsible vehicle, transportation route, destination and shipment information, load capacity, estimated time, and details of the goods transported by the vehicle. It also supports viewing detailed scheduling plans, manual editing, and exporting results. This is consistent with the result display, detailed plan viewing, failed order viewing, and export functions on the intelligent scheduling page in the manual.

[0151] (v) Step S5: Manual route change feedback, local dictionary update and local index reconstruction

[0152] Step S5 is the third high-contribution step of this invention and the key to its evolution from a static rule system to a closed-loop optimization system. The core idea of ​​this step is that the system does not immediately propagate every manual correction into a global rule. Instead, it first records the manual route-changing behavior in a structured manner as correction samples. Only when a combination of original starting and ending texts under a certain template fingerprint is repeatedly changed to the same route within a statistical window and reaches a specified consistency level is the correction considered stable. Then, only the location synonym sub-dictionary corresponding to the template fingerprint is updated, and only the local fragment of the responsibility inverted index related to the updated term is reconstructed. This approach both absorbs real business experience and avoids global mislearning caused by a few occasional corrections.

[0153] like Figure 7 As shown, step S5 can be implemented as follows.

[0154] 1. Correct sample records

[0155] When the system sends an order to the correction queue in S2 due to insufficient confidence, or generates a failed scheduling order in S4 due to insufficient resources or route ambiguity, the dispatcher can manually change the route, manually specify the target route, manually confirm the correct route, or manually edit the scheduling plan in the front-end interface. Each such operation is recorded by the system as a correction sample. A correction sample includes at least the following fields: template fingerprint, original starting point text, original ending point text, material category, original route determined by the system, manually corrected route identifier, operator, operation time, correction reason, and whether the operation was successful.

[0156] For example, if an order under template fingerprint T001 repeatedly writes "CB Warehouse West" as the starting point, and the system initially fails to stably identify the standard starting point corresponding to this text, and the dispatcher manually changes it to route R015 multiple times, then each operation will be recorded as a correction sample. Correction samples are first entered into the correction sample table and are not directly written into the location synonym sub-dictionary.

[0157] 2. Correction of Consistency Coefficient Calculation

[0158] The system can periodically scan and correct the sample table, grouping samples according to template fingerprint, original starting text, original ending text, and material category, and then statistically analyze each group of samples within a predetermined statistical window. The degree of stability within the correction. The correction consistency coefficient is preferably calculated using the following formula:

[0159] ;

[0160] in, To correct the consistency coefficient; In the statistics window Within, the number of corrected samples that have the same template fingerprint, the same original starting point text, the same original ending point text, and the same material category, and have been manually changed to the same route identifier; In the statistical window Within this range, the total number of corrected samples with the same template fingerprint, the same original starting text, the same original ending text, and the same material category; For statistical windows, such as the last 7 days, the last 15 days, or the last 30 days.

[0161] System optimization only when and Only then is it considered that the group of artificial corrections has sufficient stability. This is a threshold for the number of samples, for example, it can be 3, 4 or 5; This serves as a consistency threshold, such as 0.75, 0.80, or 0.90. Setting a dual threshold aims to prevent situations where a single, accidental correction updates the dictionary, or where an update is forced despite sufficient sample size and dispersed human opinions.

[0162] 3. Partial update of location synonym sub-dictionary

[0163] When a set of corrected samples meet the trigger condition, the system does not modify the global dictionary, but only performs a partial update on the location synonym sub-dictionary corresponding to the template fingerprint and the material category. The update methods can include the following:

[0164] First, if the original start point text or the original end point text does not exist in the sub-dictionary, a new mapping entry is added. For example, add a new entry: CB Cang Xi → Carbon Black No.1 Warehouse;

[0165] Second, if the entry already exists but the mapping target is incorrect, mark the original entry as an invalid old version and write a new version of the mapping;

[0166] Third, if the error of the text is only caused by unstable word segmentation, an intermediate standardized token can be added without directly replacing the final route result.

[0167] Preferably, when updating the dictionary, the system retains the original entry version, creation source, update time, number of trigger samples and consistency coefficient for subsequent audit and rollback. In this way, even if an update is found to be unreasonable subsequently, it can be restored to the previous version.

[0168] 4. Partial fragment reconstruction of responsibility inverted index

[0169] A further innovation of the present invention is that after the location synonym sub-dictionary is updated, it does not simply take effect temporarily when the next query is performed, but synchronously triggers the reconstruction of local fragments of the responsibility inverted index related to the updated entry. The so-called local fragment reconstruction refers to only finding the route records related to the standard start point or standard end point mapped by the updated entry, and regenerating their index fragments, instead of deleting and reconstructing all global indexes.

[0170] For example, when a new start point mapping of CB Cang Xi → Carbon Black No.1 Warehouse is added under template fingerprint T001, and Carbon Black No.1 Warehouse as a standard start point is involved in two routes R015 and R018, the system only needs to read the responsible vehicle bitmap, responsible driver bitmap, material compatibility code and return route identifier of these two routes, generate a new local index fragment and replace the old version, without reconstructing all other route indexes irrelevant to finished product warehouse, auxiliary material warehouse and the like. Through this mechanism, the update cost is significantly reduced, and the stability of route matching for other templates or other material categories will not be affected.

[0171] To ensure the consistency of partial reconstruction, the system can adopt the following steps:

[0172] First, find the standard location code mapped by the updated entry according to the entry;

[0173] Second, query all valid routes in the route base table that start or end at this standard location;

[0174] Third, retrieve the responsible vehicle and driver bitmaps from the route responsibility association table for these routes;

[0175] Fourth, read the relevant material compatibility rules and return route mapping;

[0176] Fifth, delete the corresponding local index fragments from the old version;

[0177] Sixth, write the new version's local index fragment and update the version number and effective time.

[0178] IV. Explanation of Historical Records, Output Results, and System Feasibility

[0179] like Figure 8 As shown, throughout the entire execution process of this invention, the system preferably writes the following actions uniformly into the historical record table: order import action, field parsing action, automatic route matching action, successful scheduling action, failed scheduling action, manual route rerouting action, dictionary update action, partial index reconstruction action, and result export action. Each historical record may include at least the record type, order number, original starting point text, original ending point text, target route identifier, vehicle identifier, driver identifier, reason for failure, operator, operation time, and operation result. This not only meets the traceability requirements of the enterprise's on-site operations but also facilitates subsequent verification of the source and effect of a location mapping, route resource configuration, or partial update. This function is consistent with the history record page, record details, type filtering, and export functions in the manual.

[0180] At the system implementation level, this invention can be deployed in a server or industrial control terminal environment, where the processor executes program instructions stored in the storage medium to implement the functions of the aforementioned modules. The front-end page can be accessed via a browser, the back-end service can be implemented using a conventional Web service architecture, and the database can be implemented using a combination of relational database and caching structure. Bitmap operations, index retrieval, file parsing, log recording, hash generation, etc., can all be accomplished using conventional programming techniques in the field. Software environments such as TypeScript, Vue3, Vite, NaiveUI, and Node.js described in the manual can all be considered preferred implementation environments, but they do not constitute a necessary limitation on the scope of protection of this invention.

[0181] V. Specific Application Examples

[0182] (I) Experimental Platform and Basic Data

[0183] This application example was deployed in a laboratory environment. The front-end was built using TypeScript, Vue3, Vite, and NaiveUI, while the back-end ran on Node.js 18.17. The database was MySQL 8.0, the operating system was Windows 11 Professional, the processor was an Intel Core i7-12700, and the memory was 32GB. This deployment environment exceeded the minimum hardware and software requirements specified in the manual, thus ensuring stable operation during the experiment.

[0184] The test data used in this application example is as follows:

[0185] There are 28 standard locations, including raw material warehouse, carbon black warehouse, rubber warehouse, rubber mixing workshop, calendering workshop, molding workshop, semi-finished product temporary storage area, transfer area, etc.

[0186] There are 44 fixed transportation routes;

[0187] There are 26 vehicles, including 10 eight-ton trucks, 9 ten-ton trucks, and 7 twelve-ton trucks.

[0188] 31 drivers;

[0189] The materials are categorized into four types: carbon black, adhesives, reinforcing materials, and auxiliary materials.

[0190] 1200 historical order samples were used to initialize the route responsibility semantic index, the location thesaurus, and the set of responsible vehicles / drivers;

[0191] The test consisted of 600 orders, which came from four types of order templates: procurement replenishment template, warehouse transfer template, temporary workshop transfer template, and emergency expedited template. Among them, 188 orders were non-standard text orders with location abbreviations, aliases, or abbreviations.

[0192] To verify the technical effect of the present invention, the following comparison relationship is set:

[0193] Comparative Example 1: A global thesaurus is used, but grouping is not done by template fingerprint, local feedback updates are not performed, and global vehicles and global drivers are searched during scheduling.

[0194] Example 1 - Initial Deployment: The template fingerprint grouping sub-dictionary, responsibility inverted index, and responsibility vehicle / driver restricted resource pool of the present invention are used, but feedback updates have not yet been performed.

[0195] Example 1 - Feedback Iteration: The complete scheme of this invention is adopted, and two rounds of local feedback updates are completed during the testing process.

[0196] (II) Evaluation Indicators

[0197] To quantify the effectiveness of the technology, the following evaluation metrics are used in this application example.

[0198] 1. Route automatic recognition accuracy

[0199] ;

[0200] in, To improve the accuracy of automatic route identification; The number of orders whose target routes, automatically identified by the system, match the routes ultimately confirmed by humans. This represents the total number of orders participating in the test.

[0201] 2. Manual route modification rate

[0202] ;

[0203] in, This refers to the rate of manual road rerouting. This refers to the number of orders that require manual route changes or confirmation before they can be processed. This represents the total number of orders participating in the test.

[0204] 3. Average scheduling time

[0205] ;

[0206] in, This represents the average scheduling time. For the first The computation time consumed from the time an order enters the automatic scheduling process to the time the scheduling result is output; This represents the total number of orders participating in the test.

[0207] 4. Average load factor

[0208] ;

[0209] in, Average load factor; This represents the total weight of goods actually transported within the statistical period. For the first The train number corresponds to the vehicle's rated load capacity; This represents the total number of transport trips completed within the statistical period.

[0210] 5. Empty running rate

[0211] ;

[0212] in, Empty running rate; This refers to the total mileage of vehicles traveling without load during the statistical period. This refers to the total mileage traveled by vehicles within the statistical period.

[0213] (II) Application Example 1: The effect of template fingerprint group sub-dictionary on improving route recognition accuracy

[0214] This application example is illustrated in the aforementioned figures. Figure 3 , Figure 4 and Figure 7 Please provide an explanation. Figure 3 This reflects the process of order import, template fingerprint generation, and field standardization. Figure 4 It reflects the process of starting point and ending point normalization and candidate route generation based on a location-based thesaurus. Figure 7 This reflects the closed-loop process of updating the sub-dictionary and local responsibility inverted index after manual route rerouting.

[0215] 1. Experimental Procedure

[0216] Six hundred test orders were imported into the system. Among these test orders, 188 contained non-standard location text. The system first generated a template fingerprint based on the order file's field title sequence, worksheet name, source unit identifier, and order type. Then, it called the corresponding location thesaurus based on the template fingerprint and material category to obtain a standard starting point candidate set and a standard ending point candidate set. Next, it generated a candidate route set based on the responsibility inverted index and calculated the route matching confidence. When the confidence acceptance condition was met, the target route was directly determined; otherwise, it was written to a pending correction queue for manual route modification by dispatchers on the intelligent dispatch page.

[0217] In Example 1, no distinction is made between template sources, and only a global location mapping dictionary is used; in Example 1, the template fingerprint grouping sub-dictionary is enabled on the same order; after feedback iteration, based on the previous one, the high consistency correction sample formed in the first run is written back to the corresponding sub-dictionary, and only the local fragment of the relevant responsibility inverted index is reconstructed before the second round of automatic identification and verification is performed on the same batch of test orders.

[0218] 2. Test Results

[0219] Table 1 Comparison of Route Recognition Performance

[0220]

[0221] 3. Results Analysis

[0222] From Table 1 and Figure 9As can be seen, without changing the vehicle scale, route scale, and order scale, after adopting the template fingerprint grouping sub-dictionary, the automatic route identification accuracy improved from 88.8% to 96.7%, an increase of 7.9 percentage points; the manual route correction rate decreased from 11.2% to 3.3%, indicating that the system's adaptability to location abbreviations and aliases under different templates was significantly enhanced. Furthermore, after completing two rounds of local feedback updates, the automatic route identification accuracy further improved to 98.7%, an increase of 9.9 percentage points compared to Comparative Example 1, and the manual route correction rate decreased to 1.3%, a decrease of 9.9 percentage points compared to Comparative Example 1. At the same time, the average matching time decreased from 182ms to 81ms, indicating that the sub-dictionary retrieval scope narrowed after template grouping, and the hit efficiency of the responsibility-based inverted index improved.

[0223] The results demonstrate that this invention does not rely solely on global keyword mapping. Instead, it utilizes template fingerprinting to isolate location semantics into local sub-dictionaries and then gradually corrects frequently misspelled locations through local feedback updates. This enables a more stable automatic mapping of non-standard location text in orders to target routes. For the common problem in short-distance transportation scenarios where the same location is written in multiple ways but has different meanings depending on the department and the abbreviation, this invention significantly reduces the number of misjudgments and pending confirmations.

[0224] (iv) Application Example 2: The effect of the dual-constraint resource pool of responsible vehicle / driver on improving scheduling efficiency

[0225] This application example is illustrated in the aforementioned figures. Figure 5 and Figure 6 Please provide an explanation. Figure 5 This describes the process of generating a restricted candidate resource pool based on the responsible vehicle bitmap and the responsible driver bitmap. Figure 6 This reflects the process of hierarchical scheduling in the limited candidate resource pool, which is performed in the order of direct dispatch of whole vehicles, compatible carpooling, return trip reuse, and overload splitting.

[0226] 1. Experimental Procedure

[0227] Based on the route identification completed in Example 1 - feedback iteration, automatic scheduling continues for the same batch of 600 orders. Comparative Example 1 uses a global vehicle + global driver search method, that is, each order scans available resources sequentially among all 26 vehicles and 31 drivers; this invention first reads the responsible vehicle bitmap and responsible driver bitmap according to the target route, and then performs intersection with the idle vehicle bitmap, available driver bitmap, and vehicle-driver binding relationship respectively to generate a restricted candidate resource pool, and completes the vehicle and driver assignment only within this resource pool.

[0228] 2. Test Results

[0229] Table 2 Comparison of Scheduling Efficiency and Resource Reduction Effect

[0230]

[0231] 3. Results Analysis

[0232] From Table 2, Figure 10 As can be seen, after adopting the dual-constraint resource pool for responsible vehicles / drivers of the present invention, the average number of candidate vehicles per order is reduced from 26.0 to 6.4, and the average number of candidate drivers per order is reduced from 31.0 to 7.2, significantly narrowing the resource search range; the average scheduling time is reduced from 236ms to 74ms, a reduction of approximately 68.6%. Simultaneously, the automatic scheduling success rate increases from 90.3% to 96.8%, and the manual reassignment rate decreases from 9.5% to 2.0%.

[0233] The above results demonstrate that, after the route is determined, instead of performing a coarse search on global capacity and drivers, a limited candidate resource pool is formed by the sets of responsible vehicles and drivers. This significantly reduces the search space, decreases the number of invalid attempts, and reduces subsequent manual reassignment caused by available resources that are not part of the route's regular responsibility resources. Especially in short-distance park transportation, routes, vehicle fleets, and driver teams often have relatively fixed business relationships. This invention technically represents these relationships using bitmaps and indexes, thus simultaneously improving scheduling speed and stability.

[0234] (V) Application Example 3: The effect of hierarchical scheduling mechanism on improving loading rate and empty running rate

[0235] This application example is illustrated in the aforementioned figures. Figure 6 and Figure 8 Please provide an explanation. Figure 6 It reflects the sequential processing of direct dispatch of whole vehicles, compatible carpooling, reuse of return trip vehicles, and splitting of overloaded vehicles; Figure 8 It reflects the output of scheduling results, the recording of reasons for scheduling failures, and the historical record process.

[0236] 1. Experimental Procedure

[0237] A total of 240 orders from two consecutive working days were selected for comparative testing, with a total transportation demand of 1184 tons. In Comparative Example 2, a conventional dispatching method was adopted, prioritizing direct dispatch of one vehicle per order after successful matching. Carpooling and merging were not enabled, return trip reuse was not actively activated, and overload splitting was not automatically executed. Instead, the dispatcher handled the issues manually after failures.

[0238] The present invention then executes the following sequence within the limited candidate resource pool:

[0239] (1) First determine whether the vehicle can be directly dispatched;

[0240] (2) If the optimal whole vehicle cannot be selected, determine whether carpooling can be performed according to the material compatibility rules;

[0241] (3) If a return route mapping exists and the return time meets the conditions, the return train should be reused first.

[0242] (4) If the required weight of an order is greater than the available load capacity of any candidate vehicle, it will be automatically split into multiple sub-orders and rescheduled.

[0243] 2. Test Results

[0244] Table 3 Comparison of the effects of hierarchical scheduling technology

[0245]

[0246] 3. Results Analysis

[0247] From Table 3, Figure 11 As can be seen, under the condition that the total required weight is basically the same, the solution of this invention reduces the total number of transport trips from 162 to 133, a reduction of 29 trips; the average loading rate increases from 67.8% to 84.9%, indicating that the carpooling and splitting mechanism effectively improves the fullness of single-vehicle loading; the empty-running rate decreases from 28.4% to 15.6%, indicating that the return trip reuse mechanism effectively reduces empty return trips; the number of orders successfully split automatically reaches 14, while Comparative Example 2 relies entirely on manual splitting; the number of abnormal orders requiring manual intervention decreases from 19 to 5, and the on-time completion rate increases from 95.0% to 97.9%.

[0248] This demonstrates that the present invention does not simply write several scheduling rules into the system in parallel, but rather constructs an ordered decision chain suitable for short-distance transportation, given that the target route and the limited candidate resource pool are already determined. This decision chain can automatically switch processing strategies between different order states, thereby improving overall train utilization, reducing empty running rate, and enhancing the adaptive processing capability for overloaded orders and small-batch compatible orders.

[0249] (vi) Application Example 4: The effect of local feedback update and historical tracking on improving the system's adaptive capability

[0250] This application example is illustrated in the aforementioned figures. Figure 7 and Figure 8 Please provide an explanation. Figure 7 This reflects the closed-loop process of local updates to the thesaurus of synonyms for manually rerouting sample trigger locations and local reconstruction of the inverted index of responsibility. Figure 8 It reflects the entire process of historical record keeping, failure cause tracking, and result derivation.

[0251] 1. Experimental Procedure

[0252] A total of 1920 orders were selected over four consecutive weeks for tracking and testing, approximately 480 orders per week. Initially, the system only loaded a basic location thesaurus. At the end of each week, the system analyzed the corrected samples for that week, partially updated location aliases that met the sample size threshold plus the consistency threshold, and rebuilt only the relevant responsible inverted index segments. The system recorded the accuracy of automatic route identification, the rate of manual route changes, the number of route-related failed orders, and the completeness of historical records each week.

[0253] 2. Test Results

[0254] Table 4. Continuous optimization effect after local feedback update

[0255]

[0256] 3. Results Analysis

[0257] From Table 4, Figure 12 As can be seen, after four weeks of operation, the accuracy rate of automatic route identification gradually increased from 95.2% in week 1 to 98.6% in week 4, the rate of manual route changes decreased from 6.1% to 1.9%, and the number of route-related failures decreased from 17 to 4. Meanwhile, the historical record completeness rate remained at 100%, indicating that the system can record the entire process of automatic scheduling, manual route changes, failure reasons, and result export.

[0258] This result further demonstrates that the local feedback update mechanism of the present invention has two significant advantages:

[0259] First, the system can continuously learn from manual route corrections, gradually accumulating high-frequency miswritten locations into template-level local semantic rules, thereby continuously improving the accuracy of route recognition.

[0260] Secondly, since the update scope is limited to the local synonym sub-dictionary and related responsibility inverted index fragments under the corresponding template fingerprint, it will not cause global false pollution to other templates or other material categories, thus the system is more stable.

[0261] (VII) Overall Conclusion

[0262] The above application examples and comparative experiments demonstrate that the present invention achieves at least the following technical effects:

[0263] 1. By combining template fingerprint grouping sub-dictionaries and responsibility inverted indexes, non-standard location text under multi-source order templates can be more accurately normalized into target routes, improving the automatic route recognition accuracy from 88.8% to 98.7%.

[0264] 2. By constructing a restricted candidate resource pool through the set of responsible vehicles and the set of responsible drivers, the resource search range is significantly reduced, and the average scheduling time can be reduced from 236ms to 74ms;

[0265] 3. By implementing tiered scheduling that includes direct dispatch of full vehicles, compatible carpooling, reuse of return trips, and splitting overloaded vehicles, the average load factor can be increased and the empty-run rate reduced. The average load factor can be increased from 67.8% to 84.9%, and the empty-run rate can be reduced from 28.4% to 15.6%.

[0266] 4. Through local feedback updates and local index reconstruction driven by manual route rerouting samples, the system has the ability to continuously adapt and optimize, and the historical traces are complete and traceable.

[0267] Therefore, it can be fully demonstrated that, compared with existing short-distance transportation scheduling schemes, the present invention can significantly improve route identification accuracy, shorten scheduling calculation time, improve vehicle utilization efficiency, and reduce the intensity of manual intervention, thus demonstrating outstanding beneficial technical effects.

[0268] The foregoing description of embodiments of the present invention, through which those skilled in the art are able to implement or use the present invention, will be readily apparent to those skilled in the art. Various modifications to these embodiments will be readily apparent to those skilled in the art. The general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the present invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novelty disclosed herein.

[0269] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0270] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0271] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0272] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0273] In a typical configuration, a computing device includes one or more processors (CPU), input / output interfaces, network interfaces, and memory.

[0274] Memory may include non-persistent memory in computer-readable media, such as random access memory (RAM) and / or non-volatile memory, such as read-only memory (ROM) or flash RAM. Memory is an example of computer-readable media.

[0275] Computer-readable media includes both permanent and non-permanent, removable and non-removable media that can store information using any method or technology. Information can be computer-readable instructions, data structures, modules of programs, or other data. Examples of computer storage media include, but are not limited to, phase-change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, CD-ROM, digital versatile optical disc (DVD) or other optical storage, magnetic tape, magnetic disk storage or other magnetic storage devices, or any other non-transferable medium that can be used to store information accessible by a computing device. As defined herein, computer-readable media does not include transient computer-readable media, such as modulated data signals and carrier waves.

Claims

1. A method for intelligent scheduling of short-distance transportation orders based on template fingerprint grouping feedback, characterized in that, The method includes the following steps: S1. Establish a route responsibility semantic index library, including: route base table, route responsibility association table, material compatibility rule table, return route mapping table, location synonym sub-dictionary grouped by template fingerprint and material category, and responsibility inverted index associated with the route base table; S2. Import the transportation order file and extract the order number, material category, order type, original origin text, original destination text, arrival date and required weight. Generate template fingerprints and call the corresponding location thesaurus to normalize the original origin text and original destination text to obtain the standard origin candidate set and standard destination candidate set. Then generate the candidate route set based on the responsibility inverted index. S3. When the candidate route set meets the preset matching acceptance conditions, the target route is determined, and a restricted candidate resource pool is generated by the set of responsible vehicles, the set of responsible drivers and the remaining load constraints of the vehicles corresponding to the target route; otherwise, it is written into the queue to be corrected. S4. Perform hierarchical scheduling only within the restricted candidate resource pool in the order of direct dispatch of whole vehicles, compatible carpooling, return trip reuse, and overload splitting, and output the scheduling result or the reason for scheduling failure. S5. When a user manually reroutes an order in the order to be corrected or a failed order, a correction sample is formed. When the correction sample meets the preset dictionary update trigger condition, only the corresponding location synonym sub-dictionary is updated and the relevant responsibility inverted index local segment is reconstructed.

2. The method according to claim 1, characterized in that, In step S1, the route base table records the standard starting point, standard ending point, and route identifier; the route responsibility association table records the set of responsible vehicles and the set of responsible drivers corresponding to each route identifier; the index key of the responsibility inverted index includes at least the standard starting point, standard ending point, and material category; and the index value includes at least the route identifier, the set of responsible vehicles, the set of responsible drivers, the material compatibility rule, and the return route identifier. The responsible vehicle set and the responsible driver set are stored using a responsible vehicle bitmap and a responsible driver bitmap, respectively; each index record of the responsibility inverted index also includes an available load limit field, a return route identifier field, and a material compatibility code field; wherein, the responsible vehicle bitmap is a binary bitmap representing the relationship between the responsible vehicle members for a certain route; the responsible driver bitmap is a binary bitmap representing the relationship between the responsible driver members for a certain route; The template fingerprint is generated by combining the field title sequence, worksheet name, source unit identifier, and order type of the transportation order file. It is used as the grouping key of the location thesaurus to distinguish the differences in location abbreviations, acronyms, or customary spellings under different order templates.

3. The method according to claim 1, characterized in that, In step S2, the target route is determined when the candidate route set meets the preset matching acceptance conditions; otherwise, the order is written to the queue to be corrected and transportation resources are not allocated for the time being. And / or, in step S2, the route matching confidence score is calculated for each candidate route, using the following formula: ； in, Calculate the route matching confidence score for candidate routes; The score is the match score between the original starting text and the standard starting text; The score is the score for matching the original endpoint text with the standard endpoint; A score is awarded for the consistency between the material category and the route. , and For the corresponding weights, and satisfying ; When the route matching confidence of the highest candidate route is The route matching confidence of the second-highest candidate route is At that time, only Not less than the matching confidence threshold and Not less than the confidence difference threshold In the case of [condition], the candidate route set is determined to meet the preset matching acceptance conditions.

4. The method according to claim 2, characterized in that, In step S3, the restricted candidate resource pool is generated through the following intersection operation: ； ； in, For candidate vehicle bitmaps; Location map of the responsible vehicles for the target route; This is a bitmap showing currently available vehicles. A bitmap of candidate drivers; A map showing the locations of the responsible drivers for the target route; This is a bitmap of currently available drivers; To match the candidate vehicle bitmap The driver's location map that can be bound to the vehicle; only the aforementioned and The corresponding vehicles and drivers are written into the restricted candidate resource pool.

5. The method according to claim 1, characterized in that, In step S4, the overload is split into multiple sub-orders that inherit the standard start point, standard end point, material category and arrival date of the original order, where the required weight exceeds the available load capacity of a single vehicle. And / or, in step S4, direct dispatch of the whole vehicle is executed when the required weight of a single order is not greater than the available load capacity of a single vehicle and there are no compatible orders that can be combined. Compatible carpooling is implemented when multiple orders have the same destination route, the material compatibility codes are consistent or mutually compatible, the total demand weight after merging is not greater than the available load capacity of a single vehicle, and the difference in arrival dates is not greater than the preset carpooling date window. Return trip reuse is performed when the current return route identifier of the candidate vehicle and the return route identifier of the target route are mutually mapped, and the expected return time of the candidate vehicle falls within the preset return time window; Overload splitting is performed when the order demand weight exceeds the single-vehicle available load capacity of any vehicle in the restricted candidate resource pool.

6. The method according to claim 1, characterized in that, Steps S4 and S5 further include: writing automatic dispatching operations, manual route rerouting operations, and dispatching failure operations into historical records; the historical records include at least the record type, order number, original origin text, original destination text, target route identifier, vehicle identifier, driver identifier, reason for failure, operator, operation time, and operation result; the reasons for dispatching failure include at least route not identified, insufficient responsible vehicles, insufficient responsible drivers, incompatible materials, overload, and failure to meet return conditions.

7. The method according to claim 1, characterized in that, In step S5, the template fingerprint, original starting point text, original ending point text, material category, corrected route identifier, operation time, and operation result are recorded to form a correction sample. When the correction sample corresponding to the same original starting point text, original ending point text, and material category under the same template fingerprint meets the preset dictionary update trigger condition, only the location synonym sub-dictionary corresponding to the template fingerprint is updated, and the local fragment of the responsibility inverted index associated with the updated term is reconstructed for automatic route matching and resource pool limitation of subsequent orders. And / or, in step S5, the preset dictionary update trigger condition is determined by a corrected consistency coefficient, the formula for which the corrected consistency coefficient is calculated is: ； in, To correct the consistency coefficient; In the statistics window Within, the number of corrected samples that have the same template fingerprint, the same original starting point text, the same original ending point text, and the same material category, and have been manually changed to the same route identifier; In the statistical window Within this range, the total number of corrected samples with the same template fingerprint, the same original starting text, the same original ending text, and the same material category; For statistical windows; Only when Not less than the sample size threshold and Not less than the consistency threshold Only then is the location thesaurus updated and the corresponding responsibility inverted index segment reconstructed.

8. A short-distance transportation order intelligent scheduling system based on template fingerprint grouping feedback, characterized in that, The system is used to implement the method according to any one of claims 1-7, comprising: The route responsibility semantic index module is used to establish and maintain the route base table, route responsibility association table, material compatibility rule table, return route mapping table, location thesaurus, and responsibility inverted index. The order import and template fingerprint generation module is used to import shipping order files, extract order fields, and generate template fingerprints. The route matching module is used to normalize the original starting point text and the original ending point text based on the location thesaurus corresponding to the template fingerprint and generate a candidate route set. The restricted resource pool generation module is used to generate a restricted candidate resource pool based on the set of responsible vehicles, the set of responsible drivers, the vehicle idle status, the driver available status, and the vehicle remaining load corresponding to the target route. The hierarchical scheduling module is used to perform scheduling only within the restricted candidate resource pool in the order of direct dispatch of whole vehicles, compatible carpooling, return trip reuse, and overload splitting. The feedback update module is used to record the correction samples formed by manual route changes, and when the preset dictionary update trigger conditions are met, it only updates the corresponding location thesynonym sub-dictionary and reconstructs the local fragment of the responsibility inverted index; The history tracking module is used to record information about the entire process of automatic scheduling, manual route rerouting, and scheduling failures.

9. An electronic device, characterized in that, The method includes a processor, a memory, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, it implements the method according to any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the method described in any one of claims 1 to 7.

Images