Mailbox and express cabinet priority collaborative delivery decision-making method based on package attributes

Abstract

The invention relates to the technical field of intelligent scheduling in a terminal delivery scene, in particular to a mail box and express cabinet priority collaborative delivery decision-making method based on package attributes, and the method comprises the steps: firstly obtaining a multi-dimensional attribute feature vector of a to-be-delivered package; a priority ranking result of the traditional mailbox and the intelligent express cabinet is calculated through a collaborative delivery decision function, physical state data of the two types of terminals are monitored in real time, and whether the terminals meet delivery constraint conditions or not is sequentially verified according to priorities; a delivery authorization instruction is issued to the terminal meeting the condition, if the condition is not met, the decision function is triggered to recalculate and update the priority, and meanwhile iterative optimization of the decision model is achieved based on historical delivery data. According to the invention, the resource barrier of the two types of delivery terminals is broken through, the accurate adaptation of the package demand and the terminal capability is realized, the end delivery decision accuracy and the resource utilization rate are improved, and the actual operation demand of end delivery is adapted.

Description

Technical Field

[0001] This invention relates to the field of intelligent scheduling technology in last-mile delivery scenarios, specifically a priority-based collaborative delivery decision method for mailboxes and express lockers based on package attributes. Background Technology

[0002] With the continuous and rapid development of e-commerce and express logistics, last-mile delivery, as the final link in the entire logistics service chain, directly determines the user experience and cost control level of the overall logistics service through its operational efficiency and service quality.

[0003] Smart parcel lockers have become a core infrastructure for last-mile delivery, effectively solving the industry pain point of mismatched delivery and receiving times. However, their operation generally suffers from issues such as limited locker space during peak periods and insufficient coverage in some areas. Traditional mailboxes, as standard public delivery facilities in residential communities, have a broad spatial coverage and stable physical deployment conditions, but have long suffered from limited functionality and extremely low resource utilization. The vast majority of these facilities remain idle and have failed to be effectively integrated into the modern express delivery and logistics last-mile delivery system.

[0004] Currently, in the last-mile delivery sector, the traditional mailbox and smart parcel locker delivery systems operate in isolation, lacking a unified collaborative scheduling and decision-making mechanism. Delivery personnel largely rely on personal experience and judgment in their daily deliveries, lacking a standardized and precise decision-making system based on the full range of parcel attributes. This easily leads to problems such as insufficient compatibility between parcels and delivery terminals, high delivery failure rates, and imbalanced terminal resource allocation. Most existing mainstream delivery decision-making methods are designed only for a single type of delivery terminal, failing to achieve global optimization and scheduling of resources for both types of terminals. This makes it difficult to simultaneously address the multiple core requirements of delivery efficiency, parcel security, and resource utilization efficiency, severely hindering the high-quality development of the last-mile delivery industry. Summary of the Invention

[0005] The purpose of this invention is to provide a priority-based collaborative delivery decision method for mailboxes and express delivery lockers based on package attributes, so as to solve the problems mentioned in the background art.

[0006] To achieve the above objectives, the present invention provides the following technical solution: A method for priority-based collaborative delivery decision-making between mailboxes and express lockers based on package attributes includes the following steps: Step S1: Obtain the package attribute information of the package to be delivered, and parse the package attribute information into a multi-dimensional attribute feature vector. The multi-dimensional attribute feature vector shall include at least the physical size dimension, storage security dimension, timeliness requirement dimension, and value level dimension. Step S2: Input the multi-dimensional attribute feature vector into the preset collaborative delivery decision function. The collaborative delivery decision function calculates the priority ranking results of the traditional mailbox and smart express cabinet associated with the target address based on the multi-dimensional attribute feature vector. Step S3: Real-time monitoring of the physical status data of the delivery slots of traditional mailboxes and the physical status data of the compartments of smart express lockers; Step S4: Based on the priority ranking results, verify in turn whether the physical status data corresponding to the top-ranked delivery terminals meets the delivery constraints. Step S5: If satisfied, send a delivery authorization instruction to the delivery terminal; if the currently ranked delivery terminal does not meet the delivery constraints, trigger the collaborative delivery decision function based on the multi-dimensional attribute feature vector to recalculate, generate a new priority ranking result for the remaining delivery terminals, and return to execute step S4 until an available delivery terminal is determined or a delivery failure warning is generated after traversing all terminals.

[0007] As a preferred embodiment, after generating the delivery authorization instruction or delivery failure warning in step S5, the following steps are also included: Step S6: Obtain the decision-making process data and final delivery result data generated throughout the entire delivery decision-making process, and associate and store the decision-making process data and final delivery result data in the historical delivery record database; wherein, the decision-making process data includes at least the input multi-dimensional attribute feature vector, the matching degree score and comprehensive priority score of each delivery terminal generated in the intermediate calculation, the physical status verification process record of each delivery terminal, and the list of excluded terminals; the final delivery result data includes at least the terminal identifier of the actual delivery, the delivery completion timestamp, and the final status identifier of delivery success or failure; Step S7: Extract multiple historical delivery records within a preset time period from the historical delivery record database at a preset fixed period or based on event triggering. Use the multi-dimensional attribute feature vector in each historical delivery record as the input sample, and use the corresponding actual delivery terminal identifier and delivery success status as the expected output to construct a training dataset. Step S8: Input the training dataset into the initial collaborative delivery decision function model for supervised learning and parameter optimization training. By minimizing the deviation between the predicted delivery terminal and the actual delivery terminal, dynamically adjust the weight coefficients used to calculate the matching score and the weighting factors used to fuse the availability weight coefficients within the collaborative delivery decision function. Step S9: Deploy the optimized collaborative delivery decision function model to the online decision system, replacing the original collaborative delivery decision function, and use it for priority ranking calculation of subsequent packages to be delivered, so as to realize iterative updates of the decision model based on historical delivery feedback and continuous improvement of decision accuracy.

[0008] As can be seen from the technical solution provided by the present invention above, the beneficial effects of the priority-based collaborative delivery decision method for mailboxes and express lockers based on package attributes provided by the present invention are: This invention achieves quantitative representation of core requirements across all dimensions, including physical size, storage security, timeliness, and value level, through the standardized construction of multi-dimensional attribute feature vectors of packages to be delivered. Combined with the matching degree calculation and priority ranking of the collaborative delivery decision function, it completes a deep adaptation between package delivery requirements and delivery terminal service capabilities. This model breaks away from the traditional one-size-fits-all, crude decision-making approach for last-mile delivery. It can match the most suitable delivery terminal for the personalized needs of different packages, avoiding delivery failures caused by size mismatch, substandard storage environment, or mismatched timeliness requirements from the source. This significantly improves the accuracy of single delivery decisions and effectively reduces invalid delivery operations by delivery personnel. This invention achieves efficient collaboration and global optimization of two types of last-mile delivery resources: traditional mailboxes and smart parcel lockers. Through a complete mechanism of real-time status monitoring, dynamic priority sorting, and recalculation in abnormal scenarios, it breaks down the resource data barriers between the two types of terminals. This not only fully utilizes the long-term idle delivery resources of traditional mailboxes and improves the utilization efficiency of public delivery facilities, but also effectively alleviates the operational pain point of limited locker slots during peak periods. The complementary scheduling of the two types of terminals achieves optimal allocation of last-mile delivery resources, maximizing the service capacity of existing delivery facilities. This invention constructs a closed-loop control system for the entire delivery decision-making process, ensuring high reliability and strong feasibility of delivery decisions. Through IoT links, it achieves real-time collection and availability calibration of the physical status of delivery terminals, ensuring that the terminal data used in the decision-making process is completely synchronized with the actual operating status, avoiding decision failures caused by discrepancies between historical data and actual on-site conditions. Simultaneously, through a cyclical mechanism of multi-dimensional parallel constraint verification and dynamic priority recalculation, it can quickly adjust decision-making strategies for scenarios where verification fails, maximizing the exploitation of available delivery resources within the target address, significantly improving the first-time success rate of delivery tasks, reducing the frequency of manual intervention, and effectively lowering the operating costs of last-mile delivery. This invention establishes a complete closed loop for the self-learning and self-optimization of the decision-making model, possessing strong scenario adaptability and continuous iteration capabilities. Through standardized collection and persistent storage of data throughout the entire delivery process, a high-quality training dataset can be constructed based on real historical delivery data. The core parameters of the collaborative delivery decision function are continuously optimized through supervised learning, minimizing the deviation between the model's prediction results and the actual optimal delivery results. The model can continuously adapt to the changing needs of different delivery scenarios, time periods, and package types as business data accumulates, ensuring a long-term and steady improvement in decision accuracy and fully adapting to the dynamic development needs of last-mile delivery services. This invention achieves dual benefits: cost reduction and efficiency improvement on the operational side and enhanced service experience on the user side. For high-value, time-sensitive, and high-protection packages, precise delivery can be achieved through multi-dimensional demand matching, significantly reducing the risk of package loss, damage, and overdue delivery, and fully ensuring the safety and timely fulfillment of package delivery obligations. At the same time, through intelligent collaborative decision-making and scheduling, the delivery time per package is significantly shortened, improving the overall operational efficiency of last-mile delivery. The standardized decision-making process also enables standardized management and control of last-mile delivery services, which can comprehensively improve the overall quality of last-mile delivery services and user satisfaction. Attached Figure Description

[0009] Figure 1 This is a flowchart illustrating the steps of a priority-based collaborative delivery decision method for mailboxes and express lockers based on parcel attributes, according to the present invention. Figure 2 This is a flowchart illustrating the steps following step S5 in this invention, after which a delivery authorization instruction or delivery failure warning is generated. Detailed Implementation

[0010] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0011] To better understand the above technical solutions, the following will provide a detailed explanation of the technical solutions in conjunction with the accompanying drawings and specific embodiments.

[0012] like Figure 1-2 As shown, this embodiment of the invention provides a method for priority-based collaborative delivery decision-making between mailboxes and express lockers based on package attributes, including the following steps: Step S1: Obtain the package attribute information of the package to be delivered, and parse the package attribute information into a multi-dimensional attribute feature vector. The multi-dimensional attribute feature vector shall include at least the physical size dimension, storage security dimension, timeliness requirement dimension, and value level dimension. Step S2: Input the multi-dimensional attribute feature vector into the preset collaborative delivery decision function. The collaborative delivery decision function calculates the priority ranking results of the traditional mailbox and smart express cabinet associated with the target address based on the multi-dimensional attribute feature vector. Step S3: Real-time monitoring of the physical status data of the delivery slots of traditional mailboxes and the physical status data of the compartments of smart express lockers; Step S4: Based on the priority ranking results, verify in turn whether the physical status data corresponding to the top-ranked delivery terminals meets the delivery constraints. Step S5: If satisfied, send a delivery authorization instruction to the delivery terminal; if the delivery terminal currently ranked does not meet the delivery constraints, trigger the collaborative delivery decision function based on the multi-dimensional attribute feature vector to recalculate, generate a new priority ranking result for the remaining delivery terminals, and return to execute step S4 until an available delivery terminal is determined or a delivery failure warning is generated after all terminals have been traversed. Step S6: Obtain the decision-making process data and final delivery result data generated throughout the entire delivery decision-making process, and associate and store the decision-making process data and final delivery result data in the historical delivery record database; wherein, the decision-making process data includes at least the input multi-dimensional attribute feature vector, the matching degree score and comprehensive priority score of each delivery terminal generated in the intermediate calculation, the physical status verification process record of each delivery terminal, and the list of excluded terminals; the final delivery result data includes at least the terminal identifier of the actual delivery, the delivery completion timestamp, and the final status identifier of delivery success or failure; Step S7: Extract multiple historical delivery records within a preset time period from the historical delivery record database at a preset fixed period or based on event triggering. Use the multi-dimensional attribute feature vector in each historical delivery record as the input sample, and use the corresponding actual delivery terminal identifier and delivery success status as the expected output to construct a training dataset. Step S8: Input the training dataset into the initial collaborative delivery decision function model for supervised learning and parameter optimization training. By minimizing the deviation between the predicted delivery terminal and the actual delivery terminal, dynamically adjust the weight coefficients used to calculate the matching score and the weighting factors used to fuse the availability weight coefficients within the collaborative delivery decision function. Step S9: Deploy the optimized collaborative delivery decision function model to the online decision system, replacing the original collaborative delivery decision function, and use it for priority ranking calculation of subsequent packages to be delivered, so as to realize iterative updates of the decision model based on historical delivery feedback and continuous improvement of decision accuracy.

[0013] In this embodiment, the core function of step S1 is to complete the full collection, hierarchical parsing, quantitative calculation, and standardization of the core attribute information of the entire delivery chain of the package to be delivered. This transforms the unstructured package waybill and label information into a standardized multi-dimensional attribute feature vector that can be directly used for calculating the collaborative delivery decision function. This provides a unified, accurate, and quantifiable input basis for subsequent delivery terminal priority ranking and constraint verification, ensuring the consistency and accuracy of the decision-making process. The detailed steps are as follows: Step S1-1: Collection and integrity verification of all attributes of the package to be delivered: When a delivery task is initiated, the barcode scanning device or near-field communication reading device is used to fully read the waybill information or built-in electronic tag information of the package to be delivered, and obtain the core package attribute information of the package to be delivered. The package attribute information needs to cover four types of core data: physical size parameters, storage protection mark, delivery time mark, and value level mark. At the same time, it is supplemented with two types of basic data: unique package identifier and target delivery address association information, to ensure that the collected information can fully support the subsequent delivery decision-making needs of the entire process. After information collection is completed, a completeness check of the collected information is performed. The check rule is that all four types of core attribute information must have valid values ​​and no missing or invalid fields. If the check fails, the information supplementation process is triggered until the collected information meets the completeness requirements. If the check passes, the collected full-volume packaged attribute information is structured and stored according to the preset field structure to provide a basic data source for subsequent analysis of various dimensions. Step S1-2: Physical dimension analysis and eigenvalue calculation and generation: The physical dimension parameters of the structured storage are analyzed to extract three basic values: the original measured length, width, and height of the package. To eliminate the interference of the package's placement posture on the size adaptation calculation, the package's three-dimensional dimensions are sorted in descending order. The calculation formula is as follows: ,in, The length of the package after ordered processing. The width of the package after orderly processing. The height of the package after orderly processing. The original length of the package was measured. The actual measured width of the package. The actual measured height of the package. This is a descending sorting function used to arrange the three-dimensional dimensions of packages in a fixed order from largest to smallest. Load the preset standard locker size library. This library contains the effective three-dimensional dimensions of all traditional mailbox slots and smart parcel locker slots associated with the target delivery address. Each locker size has been sorted in descending order, and the corresponding format is as follows: , , ,in, For the first The effective capacity length after the individual compartments are ordered For the first The effective accommodating width after the individual compartments are organized For the first The effective capacity height after the individual compartments are organized A unique serial number for the grid; Based on the ordered package size and compartment size, calculate the size compatibility coefficient between the package and the individual compartment. The calculation formula is as follows: ,in, For the package and the first The size adaptation coefficient of each grid opening, and the meanings of the other symbols remain consistent with the previous text; Based on the size adaptation coefficient of the full grid, the original feature value of the physical size dimension is calculated using the following formula: ,in, These are the original feature values ​​for the physical size dimension. This represents the total number of grids included in the standard grid size library; the meanings of the other symbols remain the same as above. This feature value is used to characterize the overall compatibility of the package with all delivery terminal slots associated with the target address. The higher the value, the more slots the package can be adapted to and the stronger the spatial adaptability. Step S1-3: Storage security dimension requirement analysis and feature value generation: The storage protection identifiers of structured storage are parsed field by field to identify the special storage requirements of the package. The special storage requirements cover four core scenarios: moisture protection, pressure protection, light protection, and temperature control. Each scenario corresponds to an independent identifier field. For each type of storage requirement, demand intensity quantification is performed. The quantification rule is that when there is no corresponding storage requirement, the demand intensity is 0. When there is a corresponding storage requirement, the demand intensity is divided into three levels according to the severity of the requirement: low-level demand is 0.3, medium-level demand is 0.6, and high-level demand is 0.9. After quantifying the intensity of a single type of demand, the original feature value of the storage security dimension is calculated using the following formula: ,in, To store the original feature values ​​of the security dimension, To quantify the strength required for moisture protection in packages. Quantifying the strength required for package compression protection. This is a quantification of the intensity required for light protection in packaging. Quantification of the intensity of temperature control requirements for packaging; This feature value is used to characterize the overall demand intensity of the package for the security of the delivery and storage environment. The higher the value, the more stringent the environmental protection requirements of the delivery terminal. Step S1-4: Delivery Timeliness Requirement Analysis and Feature Value Generation: Parse the delivery time stamps in the structured storage to extract the package pickup completion time and delivery deadline; obtain the current standard time of the system, and calculate the standard delivery time and remaining delivery time for the entire package process; First, calculate the standard delivery time for the entire package process. The calculation formula is: ,in, This refers to the standard delivery time for the entire process of a package, from pickup to the cut-off deadline. This is the cutoff time for package delivery. The time it takes for the package to be collected; Next, calculate the remaining delivery time for the package using the following formula: ,in, This represents the remaining delivery time for the package. This is the current standard time of the system; the meanings of the other symbols remain the same as before. Based on the remaining deliverable time and the standard delivery time throughout the entire process, the original feature value of the timeliness requirement dimension is calculated using the following formula: ,in, These are the original feature values ​​for the timeliness requirement dimension; the meanings of the other symbols remain consistent with those in the preceding text. This feature value is used to characterize the urgency of the package needing to be delivered in a timely manner. The higher the value, the stricter the time requirement for package delivery. When the value is 1, it means that the package has reached the delivery time limit and needs to be delivered immediately. Step S1-5: Package Value Level Analysis and Feature Value Generation: The value level identifier of the structured storage is parsed to extract the declared value and insured amount of the package; the maximum value between the declared value and the insured amount is taken as the effective economic value of the package. Load a preset value grading threshold system, the threshold system includes The value grading thresholds are arranged in ascending order of value. The difference between adjacent thresholds remains consistent, covering the full range of grading from the lowest to the highest value; The effective economic value of each package is matched against a value grading threshold system to determine its corresponding value grading number. The matching rule is that if the effective economic value is greater than or equal to the threshold, the package is considered a value grading number. The threshold and less than the first threshold When there are several thresholds, the corresponding classification number is: Effective economic value is greater than or equal to the highest threshold. At that time, the corresponding hierarchical number is When the effective economic value is 0, the corresponding hierarchical number is 0. Based on the matching hierarchical index, the original feature value of the value level dimension is calculated using the following formula: ,in, These are the original feature values ​​for the value level dimension. This refers to the value classification number corresponding to the effective economic value of the package. The total number of thresholds included in the value grading threshold system; This feature value is used to characterize the economic importance of the package. The higher the value, the higher the economic value of the package and the stricter the security control requirements during the delivery process. Steps S1-6: Multi-dimensional feature value normalization and feature vector generation: For the four original feature values ​​of physical size, storage security, timeliness requirements and value level, the minimum-maximum normalization process is performed to eliminate the difference in the units of feature values ​​of different dimensions, and to uniformly map the feature values ​​of all dimensions to the numerical range of 0 to 1, so as to ensure that each feature has the same numerical benchmark in subsequent decision calculations. For the original feature values ​​of a single dimension, the normalization calculation formula is as follows: ,in, The normalization result for a single-dimensional feature value. These are the original feature values ​​for this dimension. This is the preset minimum value for the feature value of this dimension. This is the preset maximum value for the feature value of this dimension; After normalizing the four feature values, the four normalized feature values ​​are sequentially combined according to a preset fixed order of physical size, storage security, timeliness requirements, and value level to generate a multi-dimensional attribute feature vector for the package to be delivered. The formula is as follows: ,in, This is a multi-dimensional attribute feature vector of the package to be delivered. This is the normalized result of the physical dimension feature values. To store the normalized results of the security dimension feature values, For the normalized result of the feature values ​​of the timeliness requirement dimension, The normalized result of the feature values ​​of the value level dimension; After the feature vector is generated, the validity of the feature vector is checked. The check rule is that all elements in the feature vector are within the range of 0 to 1, with no missing or invalid values. If the check fails, the parsing and calculation process of the corresponding dimension is re-executed. If the check passes, the generated multi-dimensional attribute feature vector is associated with the unique identifier of the package and stored, and transmitted to the subsequent collaborative delivery decision-making stage as the core input for priority ranking calculation.

[0014] In this embodiment, the core function of step S2 is to combine the multi-dimensional attribute feature vector of the package to be delivered generated in step S1 with the inherent attributes and real-time time features of the two types of delivery terminals associated with the target delivery address. Through a preset collaborative delivery decision function, the matching degree between the package and the terminal is calculated, availability weight is corrected, and comprehensive priority is quantified. Finally, a priority ranking result for traditional mailboxes and smart parcel lockers is generated, providing a core decision-making basis for subsequent verification and selection of delivery terminals. The detailed steps are as follows: Step S2-1: Collection of terminal attribute information and construction of terminal attribute vector: Based on the target delivery address corresponding to the package to be delivered, extract the terminal attribute information of all traditional mailboxes and smart parcel lockers associated with that address; the terminal attribute information needs to cover four core dimensions: compartment size and capacity, environmental security level, average idle period and historical delivery success rate, to ensure that the collected information can fully characterize the delivery adaptability and service stability of the two types of delivery terminals. After collecting terminal attribute information, standardized terminal attribute vectors are constructed for both traditional mailboxes and smart parcel lockers. For traditional mailboxes, a first terminal attribute vector is constructed, expressed by the following formula: ,in, This is the first terminal attribute vector of a traditional mailbox. This refers to the average compartment size and capacity of a traditional mailbox. The environmental security level is equivalent to that of a traditional mailbox. The average idle period of a traditional mailbox. Historical delivery success rate of traditional mailboxes; For smart parcel lockers, a second terminal attribute vector is constructed, expressed by the formula: ,in, This is the second terminal attribute vector for the smart parcel locker. This refers to the average compartment size and capacity of smart parcel lockers. For the environmental security level of smart parcel lockers, This represents the average idle period of a smart parcel locker. Historical delivery success rate of smart parcel lockers; After constructing the two types of terminal attribute vectors, a vector dimension consistency check is performed. The check rule is that the number of dimensions and the dimension order of the first terminal attribute vector and the second terminal attribute vector are completely consistent, and correspond one-to-one with the dimension order of the multi-dimensional attribute feature vectors generated in step S1. If the check fails, the process of extracting terminal attribute information and constructing vectors is re-executed. If the check passes, the two types of terminal attribute vectors are stored in a structured manner to provide basic data for subsequent matching degree calculation. Step S2-2: Parcel and delivery terminal compatibility calculation: The multi-dimensional attribute feature vector of the package to be delivered generated in step S1 is compared with the first terminal attribute vector of the traditional mailbox and the second terminal attribute vector of the smart express cabinet by cosine similarity calculation to obtain the matching score between the package and the two types of delivery terminals. First, calculate the first match score between the package to be delivered and the traditional mailbox. The calculation formula is as follows: ,in, The score represents the first match score between the parcel to be delivered and the traditional mailbox. Multidimensional attribute feature vector of the package to be delivered A function to calculate the L2 norm of a vector; Next, calculate the second matching score between the package to be delivered and the smart parcel locker. The calculation formula is as follows: ,in, This represents the second matching score between the package to be delivered and the smart parcel locker; the meanings of the other symbols remain the same as before. The matching score ranges from 0 to 1. The higher the value, the better the compatibility between the package to be delivered and the corresponding delivery terminal, and the more the terminal meets the core delivery requirements of the package. After calculating the two types of matching scores, the calculation results are associated with the corresponding terminal identifier and stored to provide basic data for subsequent priority score fusion. Step S2-3: Calculation of terminal availability weight coefficient for the current delivery period: Extract the current delivery time information of the system, load the preset historical delivery database, and extract the full historical delivery data of traditional mailboxes and smart express cabinets within the same type of time period as the current delivery time from the database; the same type of time period refers to the historical time period with the same week attribute and time period interval attribute as the current time period, to ensure that the data can accurately represent the actual availability probability of the two types of terminals within the current time period; Based on the extracted historical delivery data, the first availability weight coefficient of traditional mailboxes and the second availability weight coefficient of smart express lockers in the current time period are calculated respectively. First, calculate the first availability weight coefficient for traditional mailboxes using the following formula: ,in, This is the first availability weighting coefficient for traditional mailboxes during the current delivery period. This represents the number of times a traditional mailbox was in an idle and available state within the same historical period. The total number of status statistics for traditional mailboxes within the same historical period; Next, calculate the second availability weighting coefficient for the smart parcel locker. The calculation formula is as follows: ,in, This is the second availability weighting coefficient for smart parcel lockers during the current delivery period. This refers to the number of times a smart parcel locker was in an idle or available state within the same historical time period. The total number of times the status of smart parcel lockers is counted within the same historical period; The availability weight coefficient ranges from 0 to 1. The higher the value, the higher the probability of the corresponding delivery terminal being available in the current time period, and the stronger the certainty of successful delivery. After calculating the two types of availability weight coefficients, the calculation results are associated with the corresponding terminal identifier and stored to provide a basis for correction for subsequent priority score weighting and fusion. Step S2-4: Weighted fusion of delivery terminal overall priority scores: The matching score is weighted and fused with the availability weight coefficient of the corresponding time period to generate a comprehensive priority score for traditional mailboxes and smart express cabinets. The fusion process takes into account the compatibility of the package and the terminal, as well as the actual availability probability of the terminal in the current time period, to ensure that the priority ranking result is both adaptable and feasible. First, calculate the first overall priority score for the traditional mailbox using the following formula: ,in, This represents the first overall priority score for a traditional mailbox; the meanings of the remaining symbols remain consistent with the previous text. Next, calculate the second comprehensive priority score for the smart parcel locker. The calculation formula is as follows: ,in, This is the second comprehensive priority score for the smart parcel locker; the meanings of the other symbols remain consistent with those above. The higher the comprehensive priority score, the higher the comprehensive delivery priority of the corresponding delivery terminal, and the more suitable it is as the first choice target terminal for this delivery task. After completing the calculation of the two types of comprehensive priority scores, the calculation results are associated with the corresponding terminal identifier and stored to provide core quantitative basis for subsequent priority ranking. Step S2-5: Generation of delivery terminal priority sorting results: The first comprehensive priority score of traditional mailboxes and the second comprehensive priority score of smart express cabinets are sorted in descending order. The sorting rule is that the delivery terminal with the higher comprehensive priority score is placed in the first position in the sorting result. If the overall priority scores of the two types of delivery terminals are exactly the same, they will be sorted again according to their historical delivery success rates from high to low, with terminals having higher historical delivery success rates given priority. After sorting, the priority ranking results of the packages to be delivered for the target delivery address associated with traditional mailboxes and smart parcel lockers are generated. The ranking results must clearly indicate the order of the two types of delivery terminals, the corresponding comprehensive priority score, the matching score and the availability weight coefficient, to ensure that the ranking results are traceable and verifiable. After the priority sorting results are generated, the results are associated with the unique identifier of the package to be delivered and stored, and then transmitted to the subsequent delivery terminal status verification stage as the core execution basis for the terminal verification order.

[0015] In this embodiment, the core function of step S3 is to establish real-time data interaction between the two types of delivery terminals associated with the target delivery address through the IoT communication link. This completes the high-frequency acquisition, standardized cleaning, structured storage, and dynamic calibration of the terminal physical status data, generating a terminal dataset with real-time status identifiers. This provides accurate data support that is completely synchronized with the actual operating status for the subsequent steps of verifying delivery constraints, ensuring the executability and accuracy of delivery decisions. The detailed steps are as follows: Step S3-1: Establishment of communication link for delivery terminal and real-time acquisition of physical status data: Load the preset IoT communication interface configuration file and establish bidirectional real-time data connections with the traditional mailbox control system and the smart express cabinet control system associated with the target delivery address respectively; the communication link adopts an encrypted transmission mode to ensure the integrity and security of status data transmission; after the link is established, a heartbeat packet verification is performed. If the verification passes, the link is determined to be in an available state; if the verification fails, a reconnection mechanism is triggered until a stable and available communication link is established. The data acquisition mode is determined, and the acquisition mode is divided into two categories: fixed period sampling and event-triggered sampling. Fixed period sampling performs full state data acquisition according to the preset sampling time interval. Event-triggered sampling immediately triggers incremental data acquisition when the terminal state changes. The two modes run in parallel to ensure the real-time performance and integrity of the state data. Through the established stable communication link, the current physical status data of each delivery slot in the traditional mailbox and the current physical status data of each compartment in the smart express cabinet are obtained respectively. The physical status data of the delivery slots in the traditional mailbox covers four core data items: delivery slot opening status, locking status, idle / occupied status, and mechanical fault status. The physical status data of the compartments in the smart express cabinet covers four core data items: compartment occupancy status, cabinet door lock status, compartment idle status, and electrical control fault status. For the collected single-dimensional state data, binary quantization encoding is performed to transform the unstructured state description into a computable standardized value. For the single-dimensional state data from a traditional mailbox delivery port, the quantization encoding formula is as follows: ,in, For traditional mailbox delivery slots Quantized encoding value of item status data, The sequence number of the status data item; The quantization and encoding formula for the single-dimensional status data of the smart parcel locker compartments is as follows: ,in, For the first smart express locker compartment The quantized encoding value of the item status data, and the meanings of the other symbols remain consistent with those in the previous text; After completing a single data collection and quantization encoding, all quantized status data are associated and bound with the corresponding delivery terminal identifier, delivery port or grid unique identifier to generate the original status dataset of the single data collection, which is then transmitted to the data standardization processing stage. Step S3-2: State data cleaning, normalization, and structured storage: Data cleaning is performed on the collected raw state dataset. The cleaning process includes two core steps: outlier removal and deduplication. The outlier removal rule is that if the quantization code value of the state data exceeds the range of 0 to 1, or if there is a missing single data item in the state data, it is judged as outlier and is directly removed from the dataset. The deduplication rule is that for multiple duplicate data generated from the same delivery port or grid port and the same collection timestamp, only the most recently generated data is retained, and all other duplicate data are removed. After data cleaning is completed, the cleaned status data is normalized. For the status data of traditional mailboxes and smart express cabinets, the format is standardized according to a fixed field order of terminal identifier, cabinet or delivery port identifier, collection timestamp, and status item code value. This ensures that the status data of the two types of terminals have a unified field structure and data format, eliminating the data format differences between different terminal control systems. After format normalization, all status data is structured and stored according to a preset table structure to generate a real-time updated physical status information table. Each row of data in the physical status information table corresponds to the full status information of a delivery port or grid, and each column corresponds to a fixed field item. The data in the table is automatically updated with full or incremental updates as each collection operation is completed, ensuring that the data in the table is always synchronized with the actual operating status of the terminal. Step S3-3: Dynamic labeling of terminal availability tags and construction of real-time status dataset: Load the real-time updated physical status information table, perform a comprehensive availability status judgment for each delivery port or grid in the table, and generate a corresponding availability label; the availability label is divided into three categories: idle and available, occupied, and faulty and unavailable; First, for a single delivery slot or grid, calculate the overall usability value. The calculation formula is as follows: ,in, The overall availability value for a single delivery slot or grid. This represents the total number of status data items corresponding to the delivery port or slot. For the delivery port or compartment number Quantized encoding value of item status data; Based on the comprehensive availability judgment value and the idle occupancy indicator, dynamic marking of availability tags is performed. The marking rules are as follows: when the comprehensive availability judgment value is equal to 1 and the idle occupancy indicator is in an idle state, it is marked as idle and available; when the comprehensive availability judgment value is equal to 1 and the idle occupancy indicator is in an occupied state, it is marked as occupied; when the comprehensive availability judgment value is equal to 0, it is marked as faulty and unavailable regardless of the idle occupancy indicator status. After completing the availability tagging of all delivery ports and grid ports, the marked availability tags are associated and bound with the inherent attribute information of the corresponding delivery terminals. The inherent attribute information of the delivery terminals comes from the terminal attribute vector constructed in step S2, ensuring that the status data and attribute data of each delivery terminal correspond one-to-one. After the association and binding are completed, a real-time status dataset of the terminal with status identifiers is generated. The dataset contains four core types of information: terminal identifiers, attribute information, real-time status data, and availability tags for all traditional mailbox delivery slots and smart express cabinet slots associated with the target delivery address. The dataset is fully updated synchronously with the update of the physical status information table to ensure the real-time performance and accuracy of the data. After the real-time status dataset of the terminal is constructed, the integrity of the dataset is verified. The verification rule is that the dataset contains all grid and delivery port data of all delivery terminals associated with the target address, with no missing items, and each data item contains complete four types of core information. If the verification fails, the status data acquisition and processing process is re-executed. If the verification passes, the dataset is transmitted to the subsequent delivery constraint verification stage as the core data basis for terminal verification.

[0016] In this embodiment, the core function of step S4 is to perform multi-dimensional parallel constraint verification on the target delivery terminals based on the priority ranking result of the delivery terminals generated in step S2, combined with the real-time terminal status dataset with status identifiers constructed in step S3, according to the order of priority from high to low. This accurately filters out available terminals that fully meet the full-dimensional delivery requirements of the packages to be delivered, while clearly marking terminals that fail the verification. This provides a clear execution basis for subsequent delivery authorization or priority recalculation processes, ensuring the compliance and feasibility of delivery decisions. The detailed steps are as follows: Step S4-1: Loading priority sorting results and extracting data from the terminal to be verified: First, the core data source is loaded and verified. The priority sorting result of the delivery terminal corresponding to the package to be delivered, generated in step S2, is loaded. At the same time, the latest version of the terminal real-time status dataset with status identifiers, constructed in step S3, is loaded. A data source consistency verification is performed. The verification rule is that the delivery terminal identifier in the priority sorting result must completely match the terminal identifier in the terminal real-time status dataset, with no missing or mismatched items. If the verification fails, the real-time update process of the terminal real-time status dataset is triggered, and the status collection and dataset construction operation in step S3 is re-executed. If the verification passes, the subsequent data extraction process begins. According to the priority ranking results from high to low, extract the full data corresponding to the delivery terminal with the highest current ranking. The extracted data includes the type identifier of the delivery terminal, the inherent attribute information of the terminal, the real-time physical status data of all slots or delivery ports, and the corresponding availability label. The extracted terminal data is associated and bound with the multi-dimensional attribute feature vector of the package to be delivered generated in step S1 to form the basic dataset of a single verification task, providing unified input data for subsequent constraint verification. Step S4-2: Delivering Constraint Model Matching and Loading: Based on the type identifier of the delivery terminal to be verified, the preset delivery constraint model of the corresponding type is retrieved. The delivery terminal types are divided into two categories: traditional mailboxes and smart express cabinets. Each type of terminal corresponds to an independent delivery constraint model, and the model parameters are fully matched with the hardware characteristics and service capabilities of the corresponding terminal. The delivery constraint model includes three types of core constraint thresholds and admission rules: the compartment capacity threshold that matches the physical size of the package, the environmental state threshold that matches the package storage security requirements, and the basic admission conditions that match the current availability tag of the terminal. The three types of constraints correspond one-to-one with the dimensions of the multi-dimensional attribute feature vector of the package to be delivered, ensuring that the verification process can fully cover all the core requirements of package delivery. After loading the corresponding model, a matching check is performed between the model parameters and the attributes of the package to be delivered. The check rule is that the constraint threshold dimension in the model must be completely consistent with the dimension of the multi-dimensional attribute feature vector of the package. If the check fails, an adaptive matching update of the model parameters is triggered. If the check passes, the loaded delivery constraint model is associated with the basic dataset of the single verification task to provide a judgment standard for subsequent parallel verification. Step S4-3: Multi-dimensional parallel constraint verification execution: The multi-dimensional attribute feature vector of the package to be delivered and the real-time physical state data of the terminal to be verified are simultaneously input into the loaded delivery constraint model. The model then performs three parallel constraint verifications. The three verifications are executed synchronously without interference, ensuring verification efficiency and comprehensiveness. The first verification is the terminal basic access condition verification, used to determine whether the terminal possesses the most basic delivery access qualifications; the verification process is achieved by quantitatively calculating the terminal basic access judgment value, and the calculation formula is as follows: ,in, This is the basic access control value for the terminal. A value of 1 indicates that the basic access control verification has passed, and a value of 0 indicates that the basic access control verification has failed. The second verification is the grid size matching constraint verification, used to determine whether the terminal's physical capacity meets the space requirements of the package. The verification process is based on the ordered three-dimensional dimensions of the package and the effective capacity of the terminal's grid, calculating the size matching verification judgment value. The calculation formula is as follows: ,in, This is the verification value for grid size matching. The length of the package after ordered processing. The width of the package after orderly processing. The height of the package after orderly processing. The effective capacity length of the corresponding grid after the terminal to be verified is ordered. The effective accommodating width of the corresponding grid after it has been ordered for the terminal to be verified. This represents the effective height of the corresponding grid after the terminal to be verified is ordered. A value of 1 indicates that the size matching verification has passed, and a value of 0 indicates that the size matching verification has failed. The third verification is the storage environment adaptation constraint verification, which is used to determine whether the terminal's storage environment capabilities meet the security protection requirements of the package. The verification process is based on the storage security dimension feature value of the package and the environmental security level of the terminal, and calculates the environment adaptation verification judgment value. The calculation formula is as follows: ,in, Verification criteria for storage environment adaptation. The environmental security level of the terminal to be verified. This is the original feature value for the security dimension of the package to be delivered. A value of 1 indicates that the environment adaptation verification has passed, and a value of 0 indicates that the environment adaptation verification has failed. After all three verifications are completed, a comprehensive verification judgment result is generated. The formula for calculating the comprehensive verification judgment value is as follows: ,in, This is the comprehensive verification judgment value of the terminal to be verified. A value of 1 indicates that all three verifications have passed and the terminal meets all delivery constraints. A value of 0 indicates that at least one verification has failed and the terminal does not meet the delivery constraints. Step S4-4: Verification result determination and terminal loop verification process execution: Based on the comprehensive verification judgment value, the final judgment of the verification result is performed; if the comprehensive verification judgment value is equal to 1, it is determined that the current top-ranked delivery terminal meets all delivery constraints, the terminal is identified as the available delivery terminal for this delivery task, the current loop verification process is terminated, the verification result is associated with the terminal identifier and stored, and transmitted to the subsequent delivery authorization instruction generation stage. If the comprehensive verification judgment value is equal to 0, it is determined that the current delivery terminal to be verified does not meet the delivery constraints. In the terminal real-time status dataset, the terminal is marked as the terminal that failed the delivery verification this time, and the specific reason for the failure of the terminal is recorded. The reason for the failure corresponds to the single verification item that failed. After completing the marking and recording, the next ranked delivery terminal in the priority ranking result is extracted, and the full process verification operation from step S4-1 to step S4-4 is repeated until a delivery terminal that meets all delivery constraints is found, or all delivery terminals in the priority ranking result are traversed. After completing the full terminal traversal, if no terminal that meets the delivery constraints is found, a full terminal verification failure flag is generated. This flag is then associated with the unique identifier of the package for this delivery task and stored, and transmitted to the subsequent priority recalculation and delivery failure warning stages.

[0017] In this embodiment, the core function of step S5 is to complete the final execution and closed-loop processing of the delivery decision based on the delivery terminal constraint verification results generated in step S4; for available terminals that pass the full-dimensional verification, a delivery authorization instruction is generated and issued to complete the delivery access; for terminals that fail the verification, the exclusion list is dynamically updated and the collaborative delivery decision function is recalculated to generate an updated priority ranking result and return to the verification process, forming a cyclical decision closed loop; until all available delivery terminals that meet all requirements are locked, or after traversing all terminals associated with the target address, a standardized delivery failure warning is generated, completing the final processing of the entire delivery decision process; the detailed steps are as follows: Step S5-1: Generation and issuance of authorization commands via the terminal: Load the final verification result generated in step S4. If there is a confirmed available delivery terminal, immediately start the delivery authorization instruction generation process. During the instruction generation process, extract the unique identifier of the package to be delivered and the unique identifier of the target delivery terminal. Use the two core identifiers as the core content of the instruction. Simultaneously supplement the unique number of this delivery task and the effective time window parameter of the instruction to form a complete and compliant delivery authorization instruction. The generated delivery authorization instruction is sent to the corresponding traditional mailbox control system or smart express cabinet control system through the encrypted IoT communication interface with a stable connection established in step S3. After the instruction is sent, a response monitoring process is initiated to wait for the confirmation receipt from the terminal control system. If the confirmation receipt is received from the terminal within the preset response time window, the instruction is deemed to have been successfully sent, and the corresponding compartment or delivery port of the terminal will automatically open as required by the instruction, allowing the delivery person to complete the final delivery operation. If a valid confirmation receipt is not received within the response time window, the instruction resend mechanism is triggered. If the number of resends reaches the preset limit and still fails, the terminal is marked as a terminal that failed the delivery verification this time, and it simultaneously enters the subsequent exclusion list update and priority recalculation process. Step S5-2: Verify the failure terminal marking and dynamically update the terminal exclusion list: If step S4 generates a terminal verification failure flag, or if a terminal fails to issue a command in step S5-1, the standardized handling process for the terminal that failed verification is immediately initiated. First, the unique identifier of the terminal that failed verification and the specific reason for the verification failure or command issuance failure are extracted, and the unique identifier of the terminal is dynamically added to the exclusion terminal list specific to this delivery task. The terminal exclusion list adopts a real-time synchronous update mode. Every time a terminal that fails verification is added, the list content is updated immediately. At the same time, the reason for failure is associated with the multi-dimensional attribute feature vector of the package to be delivered generated in step S1 and stored. The stored full data is used for dynamic weight adjustment in the subsequent recalculation process to ensure that the priority ranking result after recalculation is completely consistent with the current actual available terminal resources and avoids repeated calculation and invalid verification of terminals that have been verified as unusable. Step S5-3: Recalculate the collaborative delivery decision function based on exclusion constraints: The updated exclusion terminal list is used as a constraint input, and the collaborative delivery decision function defined in step S2 is called synchronously to perform priority recalculation for the remaining available terminals. During the recalculation, the multi-dimensional attribute feature vector of the package to be delivered generated in step S1 is fully retained. At the same time, all delivery terminals that have failed verification are automatically blocked according to the exclusion terminal list, and only traditional mailboxes and smart express cabinets that have not entered the exclusion list are retained as valid candidates for recalculation. The recalculation process fully follows the standard calculation logic defined in step S2. First, the terminal attribute vector of the remaining candidate terminals is constructed. Then, the matching degree between the package and the terminal is calculated. The terminal availability weight coefficient at the current moment is updated synchronously. The weighted fusion of the matching degree score and the availability weight coefficient is completed to generate the comprehensive priority score of the remaining candidate terminals. For the availability weight coefficient during the recalculation process, a failure cause correction factor is introduced for dynamic adjustment. The calculation formula is as follows: ,in, For the revised first The availability weight coefficient of each candidate terminal The first one calculated in step S2 The original availability weight coefficients of each candidate terminal. For the first The failure reason correction factor for each candidate terminal ranges from 0 to 1. When there are no failure records of the same type, the value is 0. When there are failure records of the same type, the value increases step by step according to the number of failures. Based on the revised availability weight coefficient, the comprehensive priority score of the remaining candidate terminals is recalculated using the following formula: ,in, The first one obtained by recalculation The overall priority score of each candidate terminal, The parcel to be delivered calculated in step S2 and the first Matching scores for each candidate terminal; After calculating the comprehensive priority score of all remaining candidate terminals, sort them in descending order according to the score from high to low to generate a new priority ranking result for the remaining delivery terminals. After the ranking result is generated, a non-empty result check is performed. If there are no valid candidate terminals in the ranking result, the delivery failure warning generation process is directly entered. If there are valid candidate terminals in the ranking result, the result is synchronously transmitted to the constraint verification process in step S4. Step S5-4: Execute a loop to verify the new priority sorting results: The newly generated priority sorting result is returned to the delivery constraint verification process in step S4. The next delivery terminal to be verified is extracted in order of the updated priority from high to low, and the full-process multi-dimensional parallel constraint verification operation defined in step S4 is repeated. If, during the loop verification process, an available delivery terminal that meets all delivery constraints is found, the current loop process is immediately terminated, and the process jumps directly to step S5-1 to generate and issue the delivery authorization instruction; if, during the loop verification process, all terminals in the current sorting result fail to be verified, the process jumps back to step S5-2 to update the excluded terminal list, and simultaneously triggers a new round of priority recalculation, forming a complete loop decision-making closed loop. Step S5-5: Generation and push of delivery failure alerts after full terminal traversal: If, after multiple rounds of iterative verification and priority recalculation, all traditional mailboxes and smart parcel lockers associated with the target delivery address have been traversed and no available delivery terminal that meets all delivery constraints has been found, the current delivery decision-making process will be terminated immediately, and the standardized delivery failure warning generation process will be initiated. During the early warning information generation process, four core contents are extracted: the unique identifier of the package to be delivered, the target delivery address, the failure reasons of all terminals that failed verification, and the suggested handling methods, forming a complete delivery failure early warning information. The suggested handling methods include three standardized handling schemes: manual intervention for delivery, replanning the delivery route, and rescheduling delivery, providing clear execution guidelines for subsequent operations. The generated delivery failure warning information is simultaneously pushed to the deliveryman's handheld terminal and the back-end management system; after the push is completed, a push success receipt is generated and stored in association with the decision data of the entire process of this delivery task, completing the final closed-loop processing of the entire delivery decision process.

[0018] In this embodiment, the core function of step S6 is to complete the full collection, association, binding, structured processing, and compliant storage of all process and result data generated throughout the entire delivery decision process after the final handling of the delivery decision in step S5. This constructs a complete and traceable historical delivery record database, providing a high-quality, multi-dimensional basic data source for the subsequent training and optimization of the decision model. The detailed steps are as follows: Step S6-1: Full data collection for the entire decision-making process: After the delivery decision-making process is terminated, a full-process data collection operation is immediately initiated, collecting two core types of data: decision-making process data and final delivery result data. The decision-making process data includes the multi-dimensional attribute feature vector of the package to be delivered generated in step S1, the matching score and comprehensive priority score of each delivery terminal calculated in step S2, the real-time terminal status dataset generated in step S3, the physical status verification process record of each delivery terminal generated in step S4, the dynamically updated exclusion terminal list in step S5, and all intermediate calculation data during the recalculation of the collaborative delivery decision function. The final delivery result data includes the unique identifier of the delivery terminal actually used in this delivery task, the delivery completion timestamp, the final status identifier of successful or failed delivery, a summary of failure reasons in delivery failure scenarios, and the issuance and execution receipt data of delivery authorization instructions. After data collection is completed, all collected raw data is initially bound to the unique identifier of this delivery task to ensure that all data can be traced and located through the unique task identifier. Step S6-2: Data collection, correlation, and structuring: All collected raw data undergoes structured processing. Following a pre-defined unified data structure, all data items are standardized in terms of field format and structure. During processing, decision-making process data and final delivery result data are strongly linked and bound using a unique delivery task identifier, forming a complete data record corresponding to each delivery task. For numerical data items within the data record, format normalization is performed to ensure complete consistency in precision and dimensions for all numerical items of the same type. For non-numerical status identifier data items, binary quantization encoding is performed to convert the status description into a standardized numerical format, with the quantization encoding rules being completely consistent with the status quantization rules defined in step S3. After structured processing, a unique record identifier is generated for each delivery task's data record. This record identifier is then mapped to the unique delivery task identifier to ensure that each historical record has a globally unique location identifier. Step S6-3: Data integrity verification and storage: For each delivery record that has undergone structured processing, a data integrity check is performed. The check is achieved by calculating the data integrity score, using the following formula: ,in, For the data integrity of a single delivery record, To record the number of valid data items, This is the preset total number of required data items for a single record; The verification rule is that when the data integrity is equal to 1, the record is considered to have passed the integrity verification; when the data integrity is less than 1, the missing data completion process is triggered. After completion, the verification is re-executed until the record passes the integrity verification. Delivery records that pass the integrity verification will be automatically written to the preset historical delivery record database to complete the persistent storage of data. The historical delivery record database adopts a time-series storage structure, which is arranged in order according to the delivery completion timestamp. It supports data retrieval and extraction based on multi-dimensional conditions, providing efficient data retrieval capabilities for subsequent model training.

[0019] In this embodiment, the core function of step S7 is to use the valid historical data in the historical delivery record database to complete the selection of training samples, construction of sample pairs, and standardization processing according to the training requirements of supervised learning, ultimately forming a high-quality dataset that can be directly used for training the collaborative delivery decision function model, providing standardized input and output samples for subsequent model parameter optimization; the detailed steps are as follows: Step S7-1: Historical data extraction trigger and effective sample selection: The construction of the training dataset is triggered in two modes: fixed period trigger mode and event trigger mode. The fixed period trigger mode automatically starts the historical data extraction process at a preset fixed time interval. The event trigger mode immediately starts the historical data extraction process when the delivery failure rate exceeds a preset threshold or the decision accuracy rate declines continuously. After the process is triggered, all historical delivery records within a preset time period are extracted from the historical delivery record database. Valid samples are filtered on the extracted full historical records. The filtering rule is to retain only historical records with a final delivery status of success, remove invalid records of failed delivery, and remove abnormal records with a data integrity of less than 1, ensuring that all selected samples are valid and traceable successful delivery records. Step S7-2: Construction of supervised learning sample pairs: For the valid historical delivery records obtained through screening, sample pairs consisting of input samples and expected output samples are constructed according to the sample requirements of supervised learning; For a single valid historical record, the multi-dimensional attribute feature vector of the package to be delivered stored in the record is used as the input sample for model training; the actual delivery terminal identifier and the final status identifier of successful delivery in the record are used as the expected output sample for model training; where the actual delivery terminal identifier corresponds to the optimal priority terminal predicted by the model, and the successful delivery status identifier corresponds to the validity judgment criterion of the model prediction result. After constructing a single sample pair, a unique sample identifier is added to the sample pair, and the sample identifier is associated with the record identifier of the corresponding historical record to ensure the traceability of training samples. Step S7-3: Dataset standardization and partitioning: For the completed full set of sample pairs, perform standardization processing; for the multi-dimensional attribute feature vectors in the input samples, perform normalization processing that is completely consistent with step S1 to ensure that the numerical range and units of all input samples are completely consistent, and eliminate the interference of numerical differences between different samples on model training. After standardization, the full set of samples is divided into training and validation subsets according to a preset ratio. The training subset accounts for 80% of the full set of samples, and the validation subset accounts for 20%. The partitioning process uses stratified random sampling to ensure that the sample distribution characteristics in the training and validation subsets are completely consistent, thus avoiding the impact of data distribution bias on the model training effect. After the partitioning is completed, a standardized training dataset is generated. The training dataset includes a training subset and a validation subset, as well as corresponding sample description documents, providing complete input data for subsequent model parameter optimization training.

[0020] In this embodiment, the core function of step S8 is to input the training dataset constructed in step S7 into the initial collaborative delivery decision function model, and to perform iterative optimization of the model parameters through supervised learning, minimizing the deviation between the model's prediction results and the actual delivery results, dynamically adjusting the core weight parameters inside the model, and achieving a continuous improvement in the model's decision accuracy. The detailed steps are as follows: Step S8-1: Initial model loading and training parameter configuration: Load the initial collaborative delivery decision function model. The structure and calculation logic of this model are completely consistent with the collaborative delivery decision function defined in step S2. The model contains two types of core optimizable parameters: the dimension weight coefficient used to calculate the package and terminal matching score, and the weighting factor used to fuse the matching score and the availability weight coefficient. After loading the model, configure the core hyperparameters for the training process. The hyperparameters include the maximum number of iterations, the learning rate of the gradient descent algorithm, the number of samples in the batch, and the convergence threshold. The learning rate is set to a fixed value or an adaptive decay strategy is used, the maximum number of iterations is set to a preset fixed upper limit, and the convergence threshold is set to the minimum improvement in the accuracy of the validation set. Step S8-2: Model Prediction and Loss Function Calculation: Input samples from the training subset are batched into the initial model, which performs forward propagation calculations to generate a prediction delivery terminal priority ranking result for the corresponding input samples. The terminal identifier with the highest priority in the ranking result is extracted as the model's prediction output result. Based on the model's predicted output and the expected output corresponding to the sample, the total loss value of the model training is calculated. The total loss value is composed of a weighted average of the prediction bias loss and the regularization loss, and the calculation formula is as follows: ,in, This represents the total loss value during model training. To predict bias loss, The regularization coefficient is . This is the L2 regularization loss, used to prevent the model from overfitting. Prediction bias loss is obtained by calculating the classification cross-entropy between the model's predicted results and the expected output results. The calculation formula is as follows: ,in, This refers to the number of samples processed in a single batch. For the first The expected output result for each sample is set to 1 if the predicted terminal matches the actual terminal, and 0 if they do not. For the model to the first The predicted output probability value for each sample; The formula for calculating regularization loss is: ,in, This represents the total number of optimizable parameters within the model. For the first in the model The possible values ​​of the optimizable parameters; Step S8-3: Backpropagation and parameter iterative update: Based on the calculated total loss value, the backpropagation algorithm is executed to calculate the gradient value of the total loss value with respect to each optimizable parameter in the model. After the gradient value is calculated, the gradient descent optimization algorithm is used to iteratively update all optimizable parameters in the model. The parameter update formula is as follows: ,in, For the first The updated values ​​of the parameters For the first The values ​​of each parameter before the update. The learning rate configured for the training process. For the total loss value, the first The gradient values ​​of each parameter; After each batch of sample training is completed, a parameter update is performed; after each round of iterative training on the full training subset, the updated model's performance is evaluated using the validation subset, and the model's decision accuracy on the validation subset is calculated; the formula for calculating decision accuracy is: ,in, The decision accuracy of the model on the validation subset. To verify the number of samples within the subset whose model predictions match the expected output, To verify the total number of samples within the subset; Step S8-4: Model convergence determination and training termination: After each round of training and validation set evaluation, a model convergence check is performed. The convergence check rule is that if the improvement in validation set accuracy is less than a preset convergence check threshold after 20 consecutive rounds of training, the model is considered to have reached convergence, and the training process is immediately terminated. If the convergence check condition is not met, the next round of training iterations continues until the preset maximum number of iterations is reached. After the training process is terminated, the model parameters with the highest validation set accuracy during training are saved, an optimized collaborative delivery decision function model is generated, and all log data of the training process is saved, including the loss value of each iteration, validation set accuracy, and parameter update records, to provide data support for subsequent model performance analysis.

[0021] In this embodiment, the core function of step S9 is to deploy the optimized collaborative delivery decision function model from step S8 to the online decision system, replacing the original decision model and enabling online model updates. Simultaneously, a continuous model performance monitoring mechanism is established, continuously triggering iterative optimization of the model based on new delivery data, forming a complete closed loop of self-learning and self-optimization of the decision model, ensuring that the model's decision accuracy continuously improves with changes in the delivery scenario. The detailed steps are as follows: Step S9-1: Offline performance verification of the optimized model: After the optimized model is generated, offline performance verification is performed first. The verification process uses a test dataset independent of the training dataset, which consists of the latest historical records in the historical delivery record database that were not used in training. The test dataset is input into the optimized model and the original online model, and the three core performance indicators of decision accuracy, average decision time, and delivery success rate are calculated for the two models respectively. The verification rule is that if the decision accuracy and delivery success rate of the optimized model are both higher than those of the original online model, and the average decision time does not exceed the preset time threshold, then the model is considered to have passed the offline verification. If the verification fails, the process returns to step S7, the training dataset and training parameters are readjusted, and the model optimization training is performed again until the model passes the offline verification. Step S9-2: Model Gray-Scale Deployment and Online Validation: The optimized model, validated offline, is first deployed in a canary phase. During this phase, a predetermined proportion of new delivery tasks are diverted to the optimized model for decision-making, while the remaining tasks are handled by the original online model. The diversion ratio is gradually increased from low to high, and the online operational status of the optimized model is monitored throughout, including three core performance indicators: decision success rate, average response time, and error reporting rate. During the canary deployment, the online decision-making performance of the optimized model and the original model are compared simultaneously. If the online decision-making accuracy and delivery success rate of the optimized model both reach the predetermined targets, and the operational indicators fully meet the system requirements, the model is deemed to have passed online validation. If performance issues or operational anomalies occur, the diversion is immediately stopped, all delivery tasks are switched back to the original model, and the process returns to the model optimization phase to investigate the problem. Step S9-3: Full Model Deployment and Online Replacement: The optimized model, validated online, is then fully deployed. During full deployment, the optimized collaborative delivery decision function model is completely deployed to the online decision system. Simultaneously, the decision service of the original model is paused, and all newly added delivery tasks are switched to the optimized model for processing. After the switch is complete, full-scale operation monitoring is initiated to continuously collect the full-scale operation data and decision result data of the model, ensuring the stability of the model's full-scale operation and the absence of abnormalities in the decision process. At the same time, the optimized model parameters are synchronously backed up to the model version management library, generating the corresponding model version number and recording the model's training time, optimization content, and performance indicators, thereby achieving full lifecycle management of the model version. Step S9-4: Continuous Iterative Construction of the Closed-Loop Model After the full deployment of the model is completed, a continuous monitoring mechanism for model performance is established. The monitoring mechanism counts the model's three core indicators in real time: online decision accuracy, delivery success rate, and average decision time. If any indicator shows a continuous decline, or if the delivery failure rate exceeds a preset threshold, a new round of model training and optimization process is immediately triggered, and the entire process from steps S7 to S9 is re-executed. At the same time, according to a preset fixed cycle, the model is automatically triggered for periodic iterative optimization. Based on the latest accumulated historical delivery data, the parameters of the online model are continuously fine-tuned and the performance is optimized to ensure that the model can continuously adapt to changes in delivery scenarios. This achieves iterative updates of the decision model based on historical delivery feedback and continuous improvement of decision accuracy, forming a complete self-optimization closed loop for the model.

[0022] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for priority-based collaborative delivery decision-making between mailboxes and express lockers based on package attributes, characterized in that: Includes the following steps: Step S1: Obtain the package attribute information of the package to be delivered, and parse the package attribute information into a multi-dimensional attribute feature vector. The multi-dimensional attribute feature vector shall include at least the physical size dimension, storage security dimension, timeliness requirement dimension, and value level dimension. Step S2: Input the multi-dimensional attribute feature vector into the preset collaborative delivery decision function. The collaborative delivery decision function calculates the priority ranking results of the traditional mailbox and smart express cabinet associated with the target address based on the multi-dimensional attribute feature vector. Step S3: Real-time monitoring of the physical status data of the delivery slots of traditional mailboxes and the physical status data of the compartments of smart express lockers; Step S4: Based on the priority ranking results, verify in turn whether the physical status data corresponding to the top-ranked delivery terminals meets the delivery constraints. Step S5: If satisfied, send a delivery authorization instruction to the delivery terminal; if the currently ranked delivery terminal does not meet the delivery constraints, trigger the collaborative delivery decision function based on the multi-dimensional attribute feature vector to recalculate, generate a new priority ranking result for the remaining delivery terminals, and return to execute step S4 until an available delivery terminal is determined or a delivery failure warning is generated after traversing all terminals.

2. The method for priority-based collaborative delivery decision-making between mailboxes and express lockers based on parcel attributes as described in claim 1, characterized in that: Obtain the package attribute information of the parcel to be delivered, and parse the package attribute information into a multi-dimensional attribute feature vector. The multi-dimensional attribute feature vector includes at least the physical size dimension, storage security dimension, timeliness requirement dimension, and value level dimension, specifically including: S1-1: Obtain the package attribute information of the package to be delivered by scanning the package label or reading the electronic tag. The package attribute information shall include at least the physical size parameters, storage protection mark, delivery time mark and value level mark. S1-2: Analyze the physical size parameters, extract the length, width, and height values ​​of the package, and combine them with the preset standard compartment size library to generate feature values ​​of the physical size dimension through a size matching algorithm. These feature values ​​are used to characterize the degree of compatibility between the package and different delivery terminal compartments. S1-3: Analyze the storage protection label to identify the special storage requirements of the package, including but not limited to moisture-proof, pressure-proof, light-proof or temperature-controlled requirements. Map the identification results to feature values ​​of the storage safety dimension. These feature values ​​are used to characterize the intensity of the package's environmental safety requirements. S1-4: Parse the delivery timeliness identifier, extract the expected delivery time or deadline for delivery of the package, calculate the timeliness urgency in combination with the current time, and map it into a feature value of the timeliness requirement dimension. This feature value is used to characterize the urgency of the package needing to be delivered in a timely manner. S1-5: Analyze the value level identifier to obtain the declared value or insured amount of the package, and convert it into a feature value of the value level dimension according to the preset value level threshold. This feature value is used to characterize the economic importance of the package. S1-6: Normalize the feature values ​​of the physical size dimension, storage security dimension, timeliness requirement dimension, and value level dimension, and combine them in a preset order to generate a multi-dimensional attribute feature vector of the package to be delivered.

3. The method for priority-based collaborative delivery decision-making between mailboxes and express lockers based on parcel attributes as described in claim 1, characterized in that: The multi-dimensional attribute feature vector is input into a preset collaborative delivery decision function. Based on the multi-dimensional attribute feature vector, the collaborative delivery decision function calculates the priority ranking of the package to be delivered relative to the target address for traditional mailboxes and smart parcel lockers. Specifically, this includes: S2-1: Obtain the terminal attribute information of the traditional mailbox and smart parcel locker associated with the target address, and construct the first terminal attribute vector of the traditional mailbox and the second terminal attribute vector of the smart parcel locker respectively. The terminal attribute information includes at least the compartment size and capacity, environmental security level, average idle period and historical delivery success rate. S2-2: The matching degree of the multi-dimensional attribute feature vector is calculated with the first terminal attribute vector and the second terminal attribute vector respectively to obtain the first matching degree score between the package to be delivered and the traditional mailbox, and the second matching degree score between the package to be delivered and the smart express cabinet. S2-3: Obtain historical delivery data for traditional mailboxes and smart parcel lockers during the current delivery period, and calculate the first availability weight coefficient of traditional mailboxes and the second availability weight coefficient of smart parcel lockers during the current delivery period based on the historical delivery data. S2-4: The first matching score and the first availability weight coefficient are weighted and fused to generate the first comprehensive priority score of the traditional mailbox; the second matching score and the second availability weight coefficient are weighted and fused to generate the second comprehensive priority score of the smart express cabinet. S2-5: Sort the first comprehensive priority score and the second comprehensive priority score, and generate the priority ranking result of the traditional mailbox and smart express cabinet associated with the target address of the package to be delivered based on the ranking result.

4. The method for priority-based collaborative delivery decision-making between mailboxes and express lockers based on parcel attributes as described in claim 1, characterized in that: Real-time monitoring of the physical status data of traditional mailbox delivery slots and smart parcel locker compartments, specifically including: S3-1: Through a preset IoT communication interface, establish real-time data connections with the traditional mailbox control system and the smart express cabinet control system associated with the target address, respectively. According to the set sampling period or based on event triggering, obtain the current physical status data of each delivery slot of the traditional mailbox and the current physical status data of each compartment of the smart express cabinet. Among them, the physical status data of the delivery slot includes at least the opening and closing status, locking status, idle and occupied status, and mechanical fault status. The physical status data of the compartment includes at least the compartment occupancy status, cabinet door lock status, compartment idle status, and electrical control fault status. S3-2: Perform data cleaning and format normalization on the acquired physical status data of the delivery port and the physical status data of the grid, remove abnormal or duplicate data, and store the cleaned physical status data in a structured manner according to the delivery terminal identifier and the grid identifier to generate a real-time updated physical status information table. S3-3: Based on the physical status information table, dynamically mark the current availability tag of each traditional mailbox delivery slot and each smart express cabinet slot. The availability tag includes at least "idle and available", "occupied" or "faulty and unavailable". Associate the marked availability tag with the attribute information of the corresponding delivery terminal to form a real-time terminal status dataset with status identifiers, which is used to verify whether the delivery terminal meets the delivery constraints in subsequent steps.

5. The method for priority-based collaborative delivery decision-making between mailboxes and express lockers based on parcel attributes as described in claim 1, characterized in that: Based on the priority ranking results, the physical status data corresponding to the top-ranked delivery terminals are verified sequentially to ensure they meet the delivery constraints. Specifically, this includes: S4-1: Obtain the priority sorting result and read the terminal real-time status dataset with status identifiers. According to the priority from high to low, extract the physical status data corresponding to the traditional mailbox or smart express cabinet with the highest current ranking from the terminal real-time status dataset. S4-2: Based on the current delivery terminal type to be verified, retrieve the preset delivery constraint model. The delivery constraint model includes at least the compartment capacity threshold that matches the physical size of the package, the environmental state threshold that matches the package storage security requirements, and the basic admission conditions that match the current availability label of the terminal. S4-3: Input the multi-dimensional attribute feature vector and physical state data into the delivery constraint model at the same time. The delivery constraint model performs three parallel verifications: First, verify whether the terminal's current availability tag is in an idle and available state; second, verify whether the actual compartment size of the terminal is greater than or equal to the space requirement required by the physical size dimension of the package; and finally, verify whether the terminal's real-time environmental parameters meet the special requirements for moisture-proof, pressure-proof, or light-proof corresponding to the package storage safety dimension. S4-4: If all three verification results are passed, the top-ranked delivery terminal is determined to meet the delivery constraints and is identified as a usable delivery terminal. If any verification result is failed, the delivery terminal is determined to not meet the delivery constraints and is marked as a failed delivery terminal in the terminal real-time status dataset. At the same time, the next ranked delivery terminal in the priority ranking result is extracted, and steps S4-1 to S4-4 are repeated until a delivery terminal that meets the delivery constraints is found or all terminals are traversed.

6. The method for priority-based collaborative delivery decision-making between mailboxes and express lockers based on parcel attributes as described in claim 1, characterized in that: If the conditions are met, a delivery authorization instruction is sent to the delivery terminal; if the currently ranked delivery terminal does not meet the delivery constraints, the collaborative delivery decision function is triggered based on the multi-dimensional attribute feature vector to recalculate, generating a new priority ranking result for the remaining delivery terminals, and returning to step S4, until an available delivery terminal is determined or a delivery failure warning is generated after traversing all terminals, specifically including: S5-1: If it is determined in step S4 that the current top-ranked delivery terminal meets the delivery constraints, a delivery authorization instruction containing the identifier of the package to be delivered and the identifier of the target delivery terminal is immediately generated, and the delivery authorization instruction is sent to the corresponding traditional mailbox control system or smart express cabinet control system through the Internet of Things communication interface to open the designated compartment or delivery slot for the delivery person to complete the delivery operation. S5-2: If it is determined in step S4 that the current top-ranked delivery terminal does not meet the delivery constraints, then obtain the terminal identifier of the delivery terminal and the specific reason for its verification failure, dynamically add the terminal identifier to the excluded terminal list of this delivery task, and associate the verification failure reason with the current multi-dimensional attribute feature vector for weight adjustment in the subsequent recalculation process. S5-3: Using the exclusion terminal list as a constraint input, the collaborative delivery decision function in step S2 is called synchronously. The collaborative delivery decision function, while retaining the original multi-dimensional attribute feature vector, automatically blocks the delivery terminals that have failed verification based on the exclusion terminal list, and recalculates the matching degree and availability weight of the remaining traditional mailboxes and smart express cabinets to generate a new priority ranking result for the remaining delivery terminals. S5-4: Return the new priority sorting result to step S4, extract the next delivery terminal to be verified in the updated ranking order, repeat the verification process of physical status data until a usable delivery terminal that meets all delivery constraints is determined, and send a delivery authorization instruction to it. S5-5: If, after traversing all traditional mailboxes and smart parcel lockers associated with the target address, a delivery terminal that meets the delivery constraints is still not found, the delivery decision process is terminated, and a delivery failure warning message containing the package identifier, the reason for failure, and the suggested handling method is automatically generated. This warning message is pushed to the delivery terminal or the back-end management system to prompt manual intervention or replanning of the delivery route.

7. The method for priority-based collaborative delivery decision-making between mailboxes and express lockers based on parcel attributes as described in claim 1, characterized in that: After generating the delivery authorization instruction or delivery failure warning in step S5, the following steps are also included: Step S6: Obtain the decision-making process data and final delivery result data generated throughout the entire delivery decision-making process, and associate and store the decision-making process data and final delivery result data in the historical delivery record database; wherein, the decision-making process data includes at least the input multi-dimensional attribute feature vector, the matching degree score and comprehensive priority score of each delivery terminal generated in the intermediate calculation, the physical status verification process record of each delivery terminal, and the list of excluded terminals; the final delivery result data includes at least the terminal identifier of the actual delivery, the delivery completion timestamp, and the final status identifier of delivery success or failure; Step S7: Extract multiple historical delivery records within a preset time period from the historical delivery record database at a preset fixed period or based on event triggering. Use the multi-dimensional attribute feature vector in each historical delivery record as the input sample, and use the corresponding actual delivery terminal identifier and delivery success status as the expected output to construct a training dataset. Step S8: Input the training dataset into the initial collaborative delivery decision function model for supervised learning and parameter optimization training. By minimizing the deviation between the predicted delivery terminal and the actual delivery terminal, dynamically adjust the weight coefficients used to calculate the matching score and the weighting factors used to fuse the availability weight coefficients within the collaborative delivery decision function. Step S9: Deploy the optimized collaborative delivery decision function model to the online decision system, replacing the original collaborative delivery decision function, and use it for priority ranking calculation of subsequent packages to be delivered, so as to realize iterative updates of the decision model based on historical delivery feedback and continuous improvement of decision accuracy.

Images