Method for optimizing configuration and scheduling of multi-modal distribution resources of online drug platform

Abstract

The application discloses a medicine online platform multimode distribution resource optimization configuration and scheduling method, relates to the computer technical field, and specifically comprises the following steps: firstly, a unified data platform is constructed, and multi-source data is integrated; secondly, a three-level distribution area model is established by adopting an entropy dynamic grid division algorithm, and an adaptation matrix is constructed according to medicine and distribution mode types; thirdly, a scheduling strategy is trained based on reinforcement learning; and finally, grid division and the scheduling strategy are continuously optimized through a closed-loop feedback mechanism, so that efficient and safe medicine distribution is realized; the multi-target collaborative scheduling and closed-loop iterative optimization mechanism based on reinforcement learning significantly improve the distribution resource utilization efficiency and order processing capacity, effectively guarantee the transportation safety and time efficiency stability of various medicines, the method can intelligently respond to complex and changeable real-time scenes, realizes the comprehensive optimization of cost, time efficiency, safety and service experience, and overall improves the operation benefit and comprehensive quality of the distribution service.

Description

Technical Field

[0001] This invention relates to the field of computer technology, specifically to a method for optimizing the allocation and scheduling of multimodal delivery resources for online drug delivery platforms. Background Technology

[0002] With the rapid development of online pharmaceutical services, the timeliness, safety, and resource utilization efficiency of drug delivery have become core concerns for the industry. The multi-source data accumulated by online platforms covers multiple dimensions such as orders, drug attributes, traffic conditions, and delivery resources, providing a solid data foundation for optimizing delivery resources. Drugs themselves have diverse storage conditions, regulatory levels, and transportation risks, and delivery methods also need to be differentiated according to timeliness requirements and safety control standards. This requires the establishment of a scientific regional division system and a drug-delivery method adaptation mechanism. At the same time, as the urban delivery scope continues to expand and last-mile delivery demand becomes increasingly dispersed, the coordination difficulty of cross-regional planning and last-mile execution is constantly increasing. The scheduling model that relies solely on manual experience or fixed processes can no longer meet the needs of industry development. The mature application of technologies such as big data processing, algorithm models, and reinforcement learning provides technical support for solving the problem of delivery resource allocation, promoting the industry's transformation towards precise, dynamic, and efficient delivery scheduling. There is an urgent need for a complete solution that integrates data, regional division, adaptation rules, and dynamic scheduling.

[0003] Traditional online drug delivery resource allocation and scheduling methods have many obvious limitations. In terms of data management, various types of data are stored in different systems, lacking a unified big data service platform. Data is difficult to synchronize and integrate in real time, resulting in a lack of comprehensive and accurate data support for scheduling decisions, which can easily lead to judgment biases. The division of delivery areas often adopts a fixed model, failing to fully consider the dynamic changes in order distribution, traffic conditions, and drug demand. This causes the division of areas to be out of touch with actual delivery needs, resulting in either idle resources in some areas or insufficient transportation capacity in others. The matching of drugs and delivery methods lacks systematic and clear rules, and the matching logic is vague, making it difficult to meet the special transportation requirements of different types of drugs. The safety management of high-risk drugs cannot be adequately guaranteed. The formulation of scheduling strategies often focuses on a single objective, failing to comprehensively balance multiple core dimensions such as cost, timeliness, safety, and customer satisfaction. Moreover, it lacks a dynamic adjustment mechanism and cannot be optimized in a timely manner based on actual delivery results. It has a weak ability to cope with order fluctuations and traffic emergencies, and it is difficult to adapt to complex and ever-changing delivery environments. Summary of the Invention

[0004] The purpose of this invention is to overcome the shortcomings of existing technologies and provide a method for optimizing the allocation and scheduling of multimodal delivery resources on online drug platforms. This method integrates multi-source heterogeneous data by constructing a data platform, dynamically generates a three-level delivery area model based on an entropy-driven grid partitioning algorithm, establishes a drug-delivery mode adaptation matrix, and uses an entropy-driven reward algorithm combined with reinforcement learning to train an intelligent scheduling strategy. Finally, through a closed-loop feedback mechanism, it achieves continuous collaborative optimization of grid partitioning and scheduling strategies, thereby improving the efficiency, safety, and satisfaction of drug delivery.

[0005] To solve the above-mentioned technical problems, this invention provides the following technical solution: a method for optimizing the allocation and scheduling of multimodal delivery resources for online drug platforms, the specific steps of which are as follows: S100, Data Foundation Construction: Build an online big data resource service platform for pharmaceuticals, collect multi-source heterogeneous data, construct a unified dataset after data preprocessing, store the data and complete real-time data synchronization and on-demand retrieval; S200, Dynamic Grid Partitioning: Adopts the entropy-driven grid partitioning algorithm, constructs an initial delivery area grid based on real-time data from the big data resource service platform, dynamically adjusts the grid size and generates a three-level delivery area model, and synchronizes it to the big data resource service platform; S300, Adaptation Matrix Construction: Based on the data of the big data resource service platform, drug types and delivery method types are classified, adaptation rules are established, a drug-delivery method adaptation matrix is ​​constructed, and the matrix is ​​dynamically iterated by combining the entropy dynamic grid partitioning algorithm with the feedback data, and stored in the big data resource service platform; S400, Scheduling strategy training: The entropy-driven state reward algorithm is adopted. Based on the three-level delivery area model, the state space and action space of reinforcement learning are constructed. Combined with the drug-delivery mode adaptation matrix constraint, the scheduling strategy is generated and synchronized to the big data resource service platform. S500, closed-loop iterative optimization: Deploy a real-time feedback closed-loop iterative mechanism, collect actual delivery result data to calculate the indicator deviation rate, dynamically adjust relevant weights, coordinate the entropy-driven grid partitioning algorithm and the entropy-driven reward algorithm to iteratively optimize the grid partitioning results, and update the scheduling strategy.

[0006] Furthermore, in S100, during the data foundation construction, the multi-source heterogeneous data includes order data, drug attribute data, traffic condition data, delivery resource data, customer evaluation data, and actual delivery result data. Specifically, order data includes delivery address, order quantity, and order time; drug attribute data includes drug category and storage requirements; traffic condition data includes road conditions and real-time traffic conditions; delivery resource data includes delivery personnel and delivery equipment status; customer evaluation data includes evaluation content and evaluation level; and actual delivery result data includes delivery completion time and delivery status. The data is collected from corresponding record channels: order data is collected from transaction records, drug attribute data from drug archives, traffic condition data from real-time traffic feedback channels, delivery resource data from delivery-related records, customer evaluation data from evaluation submission records, and actual delivery result data from delivery completion feedback. During the collection process, each type of data undergoes preliminary screening to remove obviously invalid data.

[0007] Furthermore, in S200, during dynamic mesh partitioning, the mathematical expression for the entropy-driven mesh partitioning algorithm is: in, Divide the dynamic grid into entropy coefficients; Order density spatiotemporal entropy measures the degree of dispersion of orders in time and space within a grid. The spatiotemporal entropy of traffic congestion measures the degree of fluctuation of traffic status within a grid in time and space. The spatiotemporal entropy of drug demand measures the degree of demand concentration for cold chain or specially managed drugs within a grid. For time The decay factor changes and is dynamically adjusted over time; , , For dynamic weights.

[0008] Furthermore, in S200, during the dynamic grid division, the initial delivery area grid is a fixed-size square grid covering the entire target delivery service area, with an initial size of 1000 meters × 1000 meters. Each grid is assigned a unique grid number and bound to corresponding geographical latitude and longitude coordinates. The initial delivery area grid is constructed as follows: based on the geographical boundary data of the target delivery service area provided by the big data resource service platform, an initial grid for the entire area is generated using an equidistant division method. Each grid is assigned a unique grid number and bound to corresponding geographical latitude and longitude coordinates. Then, the initial grid is verified by combining the platform's historical order distribution data and traffic network data, and invalid grids without delivery needs are eliminated. The size, number, and coordinate information of the verified initial grid are stored in the grid model library of the big data resource service platform as the basic benchmark for dynamically adjusting the grid size.

[0009] Furthermore, in S200, the three-level delivery area model in the dynamic grid partitioning is a multi-level delivery area system constructed based on dynamically adjusted grids. The structure includes three levels: the first level is the city-level delivery area, which is divided according to the city's administrative geographical boundaries, covering the entire delivery geographical area and carrying the functions of cross-regional scheduling and resource coordination; the second level is the area-level delivery area, which is divided according to the city's business districts and traffic arteries, covering the main delivery concentration areas within the city and carrying the functions of delivery task allocation and capacity allocation within the area; the third level is the grid-level delivery area, which is composed of dynamically adjusted grids, covering the last-mile delivery demand points and carrying the functions of last-mile delivery execution and demand response. The generated three-level delivery area model is synchronized to the big data resource service platform as the geographical basis for subsequent delivery scheduling.

[0010] Furthermore, in S300, the main content of classifying drug types and delivery methods in the adaptation matrix construction is as follows: Based on drug storage conditions, regulatory management level, and transportation risk level, drugs are classified into three categories: cold chain drugs, specially managed drugs, and ordinary room temperature drugs. Cold chain drugs include biological agents and vaccines, which require low-temperature storage throughout the entire process and real-time temperature control during transportation. Specially managed drugs include narcotic drugs and psychotropic drugs, which must comply with special control regulations. Ordinary room temperature drugs include oral and external preparations that do not require special storage and transportation conditions. When classifying delivery methods, based on delivery time requirements, transportation conditions, and safety control level, delivery methods are divided into five categories: instant delivery, half-day delivery, next-day delivery, dedicated cold chain delivery, and dedicated personnel delivery for special drugs. Instant delivery corresponds to delivery needs within 30 minutes, half-day delivery corresponds to delivery needs within 4 hours, next-day delivery corresponds to delivery needs within 24 hours, dedicated cold chain delivery corresponds to delivery needs requiring low-temperature control throughout the entire process, and dedicated personnel delivery for special drugs corresponds to delivery needs requiring closed-loop management by dedicated personnel. The results of these two classifications serve as the basic input for establishing adaptation rules.

[0011] Furthermore, in the S300 adaptation matrix construction, the adaptation rules are based on the requirements of the drug type and combined with the functional characteristics of the delivery method. The specific content of the adaptation rules is as follows: Cold chain drugs must be adapted to dedicated cold chain delivery, meeting the requirement of 2-8℃ low-temperature control throughout the entire process. Temperature control data must be uploaded in real time during delivery to ensure that the drug storage conditions do not deviate from the standard; Specially managed drugs must be adapted to dedicated personnel for special drug delivery, requiring certified personnel to escort the drugs throughout the process. Closed-loop management is implemented during delivery, recording the escort personnel's trajectory and the drug's status throughout the process, and strictly prohibiting intermediate transfer and illegal storage; Ordinary room temperature drugs... Medicines can be matched with instant delivery, half-day delivery, and next-day delivery. The matching is determined based on the order's delivery time requirements. Instant delivery should prioritize matching with nearby delivery resources, half-day delivery should coordinate regional transit resources, and next-day delivery should be rationally planned in conjunction with trunk line delivery resources. The matching rules explicitly prohibit unsuitable combinations across different types. At the same time, supplementary constraints are made based on the risk level of medicine transportation. Medicines with high risk levels need to have higher delivery control standards. The matching rules serve as the core basis for constructing the medicine-delivery method matching matrix, ensuring that the matrix matching results comply with medicine transportation regulations and delivery time requirements.

[0012] Furthermore, in S400, during the training of the scheduling strategy, the formula for the entropy-driven reward algorithm is: in, The dynamic reward value is the comprehensive reward result output by the entropy-driven reward algorithm, used to evaluate the quality of the scheduling strategy. The higher the value, the better the adaptability and effect of the corresponding scheduling strategy. , , , The weighting is dynamic, adjusted based on the deviation rate of actual delivery results. When a certain indicator shows risk or deviation, the corresponding weight will dynamically increase. When there is a risk to drug safety, the weighting will automatically increase. Weights are assigned to reinforce reward constraints in this dimension. As a delivery cost incentive, the lower the unit order cost, the higher the reward value, which is used to guide scheduling strategies to optimize cost control; As a delivery timeliness reward, the smaller the difference between the actual delivery time and the expected time, the higher the reward value, which is used to guide the scheduling strategy to improve the stability of delivery timeliness; The higher the cold chain temperature compliance rate and the drug integrity rate, the higher the reward value, which is used to ensure the safety management of the drug transportation process. The reward is for customer satisfaction; the higher the user rating and repurchase rate, the higher the reward value, which is used to improve the terminal service experience. The grid entropy reward weight reflects the contribution of grid partitioning quality to scheduling performance; The dynamic grid partitioning adaptation index is obtained from the entropy dynamic grid partitioning algorithm and is used to quantify the degree of adaptation between the grid partitioning and the current delivery needs.

[0013] Furthermore, in S400, the specific content of the scheduling strategy training is as follows: Based on the state of each level of the three-level delivery area model, a reinforcement learning state space is constructed, including the resource reserve of the city-level delivery area, the order density of the urban area-level delivery zone, and the capacity distribution of the grid-level delivery area; an action space is constructed, including order dispatch rules, cross-regional resource allocation paths, and last-mile delivery priority settings. Combined with the constraints of the drug-delivery method adaptation matrix, it is determined that cold chain drugs are only matched with cold chain-specific delivery resources, specially managed drugs are only matched with dedicated personnel delivery resources, and ordinary room temperature drugs are matched with time-appropriate delivery methods as needed. An entropy-driven dynamic reward algorithm is adopted, and the scheduling action is evaluated through the dynamic reward value D-MOR. When the drug safety risk increases, the weight of the safety dimension is automatically increased to guide the strategy to prioritize the transportation of high-risk drugs; combined with the grid adaptation degree fed back by the entropy-driven grid partitioning algorithm, the capacity allocation within the region is dynamically adjusted.

[0014] Furthermore, in the S500 closed-loop iterative optimization, the actual delivery result data includes multi-dimensional core operational data. The specific content and collection methods are as follows: Delivery cost data includes unit order delivery cost and delivery resource idle rate, automatically generated by the big data resource service platform from order settlement data and resource scheduling logs; Delivery timeliness data includes actual delivery time and the proportion of overdue orders, transmitted in real-time by GPS positioning of the delivery terminal, automatically comparing with the expected delivery time to calculate the deviation; Drug safety data includes cold chain temperature compliance records, drug integrity rate, and special drug escort trajectory, automatically uploaded by cold chain transportation equipment with full-process temperature control data, and delivery personnel taking photos of the intact state of drugs through terminals, with special drugs relying on electronic control tags to record the closed-loop escort trajectory; Customer satisfaction data includes user ratings and complaint rates, generated by user-submitted evaluation information from the user-end APP and synchronized complaint records from customer service; Grid adaptation data includes order completion rate and resource matching saturation for each grid, dynamically matched with resources within the grid in real-time by the entropy-driven grid partitioning algorithm.

[0015] Compared with existing technologies, this method for optimizing the allocation and scheduling of multimodal delivery resources on an online drug platform has the following advantages: I. This invention establishes a big data resource service platform that integrates multi-source heterogeneous data, constructs a unified dataset, and enables real-time synchronization and on-demand retrieval. This provides data support for optimizing delivery resources. Based on the entropy-driven grid partitioning algorithm, it dynamically adjusts the grid size by combining multi-dimensional spatiotemporal characteristics such as orders, traffic, and drug demand, generating a three-level delivery area model. This forms a hierarchical and dynamically adaptable delivery geographic system, breaking the limitations of fixed area divisions. Adaptation rules are established based on precise division of drug types and delivery methods, constructing a dynamically iterative drug-delivery method adaptation matrix. This ensures the scientific matching of drugs with different attributes and delivery resources, avoids risks caused by incompatible combinations, and improves delivery compliance and resource utilization efficiency. Through a data-driven area division and resource matching mechanism, it solves the problems of rigid area division and unreasonable resource matching in traditional delivery, ensuring basic adaptability and standardization during drug transportation.

[0016] II. This invention constructs a state and action space for reinforcement learning through an entropy-driven reward algorithm. It incorporates multi-dimensional indicators such as delivery cost, timeliness, safety, and customer satisfaction into the reward evaluation system. Weights are dynamically adjusted based on grid adaptability, guiding the scheduling strategy towards multi-objective optimization. This clarifies the matching rules for delivery resources for different types of medicines, strengthens the transportation control of high-risk medicines, and ensures stable transportation safety and timeliness. Relying on a closed-loop iterative optimization mechanism, it collects multi-dimensional actual delivery result data and continuously optimizes the grid partitioning algorithm and reward algorithm, responding in real-time to changes in variables such as orders, traffic, and drug demand. This collaborative optimization model improves the utilization rate of delivery resources and order processing efficiency, reduces the risk of delays, and ensures the safety of drug transportation, forming a virtuous cycle of delivery optimization. This significantly improves the overall quality and operational efficiency of the delivery service and adapts to complex and ever-changing delivery scenarios.

[0017] Other advantages, objectives and features of the invention will be set forth in part in the description which follows, and in part will be apparent to those skilled in the art from the following examination or study, or may be learned from the practice of the invention. Attached Figure Description

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

[0019] Figure 1 A flowchart illustrating the multimodal delivery resource optimization and scheduling method for online drug delivery platforms; Figure 2A data transmission diagram illustrating a method for optimizing the allocation and scheduling of multimodal delivery resources on an online drug delivery platform; Figure 3 This is a flowchart illustrating the data flow and constraint adaptation process for the scheduling strategy of this invention. Detailed Implementation

[0020] To further illustrate the technical means and effects of the present invention in achieving its intended purpose, the following detailed description of the specific implementation methods, structures, features, and effects of the present invention, in conjunction with the accompanying drawings and preferred embodiments, is provided below.

[0021] Example 1: Example of Omnichannel Delivery by a Large-Scale First-Tier City Pharmaceutical E-commerce Platform This embodiment applies to a large-scale pharmaceutical e-commerce platform in a densely populated, high-volume, and diverse city with a wide variety of medicines. The platform needs to cover cross-regional delivery and precise last-mile delivery within the entire city's administrative area, adapting to the transportation needs and varying timeliness requirements of different medicines. Through standardized and dynamic resource allocation and scheduling throughout the entire process, it addresses the multiple challenges posed by the complex urban delivery environment. Figure 1 As shown.

[0022] A pharmaceutical online big data resource service platform was established as the core data support for the entire delivery system. It comprehensively collects multi-source heterogeneous data to ensure data coverage of key links in the entire delivery chain. Order data is accurately collected from the platform's transaction records, covering delivery addresses, order quantities, and order times across all areas of the city. By integrating this data, the distribution patterns of orders in different areas and time periods can be grasped in real time, providing a precise basis for subsequent regional division and capacity allocation. Drug attribute data is taken from the platform's drug archives, clearly defining the categories and storage requirements of various drugs. This data dimension directly determines the matching direction of subsequent delivery methods, avoiding delivery risks caused by a lack of understanding of drug characteristics. Traffic condition data is collected from real-time urban traffic feedback channels, covering the traffic conditions and real-time road conditions of major roads. Real-time updated traffic data helps... The dispatch system avoids congested road sections, ensuring the stability of delivery timeliness. Delivery resource data comes from platform-related records, including the on-duty status of delivery personnel, and the operation of delivery vehicles and cold chain equipment. Clear resource data ensures that the system has a clear understanding of the available and incremental capacity during dispatching, avoiding resource idleness or capacity gaps. Customer evaluation data is collected from user evaluation submission records, including evaluation content and rating. This feedback information directly reflects the weaknesses in delivery services, providing direction for subsequent optimization. Actual delivery result data is obtained from delivery completion feedback, covering delivery completion time and delivery status, providing basic data support for closed-loop iteration. During the collection process, various types of data undergo preliminary screening to remove obviously invalid data, ensuring that the information entering the dataset is authentic and valid. Based on this, a unified dataset is constructed and stored in the platform's dedicated data storage module, with a real-time synchronization mechanism established to enable on-demand data retrieval at each stage. This process completely breaks down the barriers of previous scattered data storage and data incompatibility between systems, allowing all subsequent decision-making stages to be based on comprehensive, real-time, and accurate data, ensuring the scientific nature of delivery resource allocation and dispatch from the source.

[0023] Based on the city's administrative and geographical boundary data provided by the big data resource service platform, an initial delivery area grid covering the entire region was generated using an equidistant division method. Each grid was assigned a unique number and bound to corresponding geographical latitude and longitude coordinates. This design enables precise positioning and management of the delivery area, ensuring that each delivery node can be clearly identified. Subsequently, the initial grid was validated by combining historical order distribution data and urban traffic network data from the platform, eliminating invalid grids with no delivery needs, effectively reducing the waste of subsequent computing resources. The size, number, and coordinate information of the validated initial grid were stored in the grid model library of the big data resource service platform, providing a basic template for subsequent dynamic adjustments. Next, the entropy-driven grid partitioning algorithm was adopted. The mathematical expression of the entropy-driven grid partitioning algorithm is: in, Divide the dynamic grid into entropy coefficients; The order density is the spatiotemporal entropy; The spatiotemporal entropy of traffic congestion; The spatiotemporal entropy of drug demand; For time The changing attenuation factor; , , The system uses dynamic weighting, combining real-time order density, traffic congestion, and drug demand data from the platform to dynamically adjust grid size. This ensures the grid division accurately adapts to the changing delivery environment, avoiding the disconnect between fixed grids and actual demand. Based on the dynamically adjusted grid, a three-tiered delivery area model is constructed. The city-level delivery area covers the entire city's administrative area, responsible for cross-regional resource coordination and scheduling. By coordinating citywide transportation capacity, it effectively balances capacity gaps in different areas, preventing imbalances where some areas have excess resources while others lack sufficient capacity. The district-level delivery area is divided based on the city's core business districts and major transportation routes, covering core areas with concentrated orders within the city. It is responsible for allocating delivery tasks and allocating transportation capacity within the district, enabling rapid response to order fluctuations and improving task allocation efficiency. The grid-level delivery area, composed of dynamically adjusted grids, precisely covers last-mile delivery demand points such as communities and office buildings, undertaking last-mile delivery execution and rapid response functions, shortening last-mile delivery distances and reducing delivery delays. The generated three-level delivery area model is synchronized to the big data resource service platform, providing a structured regional foundation for subsequent adaptation matrix construction and scheduling strategy training, enabling the entire delivery system to form a complete hierarchy from macro-level planning to micro-level execution, thereby improving overall operational efficiency.

[0024] Leveraging drug attribute data and distribution resource data from a big data resource service platform, a refined classification of drug types and distribution methods is conducted. Based on drug storage conditions, regulatory management levels, and transportation risk levels, drugs sold on the platform are categorized into three types: cold chain drugs, specially managed drugs, and ordinary room temperature drugs. This clear drug classification clarifies the core distribution requirements of different drugs, providing a clear direction for subsequent adaptation. Based on delivery timeliness requirements, transportation conditions, and safety control levels, distribution methods are divided into five categories: immediate delivery, half-day delivery, next-day delivery, dedicated cold chain delivery, and dedicated personnel delivery for special drugs. This diversified distribution method classification can meet the differentiated needs of various drugs. Based on the demand for different types of medicines and combined with the functional characteristics of delivery methods, scientific adaptation rules are established. Cold chain medicines are adapted to dedicated cold chain delivery to ensure that medicines remain in a suitable temperature environment during transportation, preventing deterioration and ineffectiveness due to temperature fluctuations. Specially managed medicines are adapted to dedicated personnel for delivery, strictly adhering to regulatory requirements to ensure compliance and safety during transportation and eliminate risks during circulation. Ordinary room temperature medicines are adapted to immediate, half-day, or next-day delivery based on order timeliness requirements, fully meeting the timeliness expectations of different users and improving user experience. At the same time, incompatible combinations are explicitly prohibited and supplemented with risk constraints to avoid unreasonable delivery arrangements from a systemic perspective. Combining the medicine demand and delivery resource distribution data within the grid fed back by the entropy-driven grid partitioning algorithm, the adaptation matrix is ​​dynamically iterated, allowing the adaptation rules to fit the resource characteristics and demand structure of different regions, avoiding a "one-size-fits-all" adaptation model. After being stored in the big data resource service platform, it provides a clear matching basis for subsequent scheduling strategies, ensuring that each order receives the most suitable delivery plan, maximizing the quality of delivery services while ensuring safety and compliance.

[0025] Based on a three-tiered delivery area model, a reinforcement learning state space and action space are constructed. The state space includes the resource reserves of cold chain equipment and dedicated delivery personnel in city-level delivery areas, the real-time order density in area-level delivery areas, and the distribution of delivery personnel and vehicle capacity in grid-level delivery areas, comprehensively covering the core influencing factors in the delivery process and enabling scheduling decisions to be based on a precise grasp of the overall state. The action space covers order dispatch rules, cross-area resource allocation paths, and last-mile delivery priority settings, providing diverse execution solutions for scheduling. Combined with the constraints of the drug-delivery method adaptation matrix, it is clarified that cold chain drugs are only matched with dedicated cold chain delivery resources, specially managed drugs are only matched with dedicated delivery personnel, and ordinary room-temperature drugs are matched with corresponding time-sensitive delivery methods as needed, preventing incompatible delivery from the execution level. An entropy-driven reward algorithm is adopted, the mathematical expression of which is: in, This is a dynamic reward value; , , , Dynamic weights; As a reward for delivery costs; Incentives for timely delivery; Rewards for drug safety; Rewards for customer satisfaction; Assign grid entropy reward weights; The algorithm assigns an entropy coefficient to the dynamic grid, comprehensively considering core dimensions such as delivery cost, timeliness, drug safety, and customer satisfaction. It evaluates scheduling actions through dynamic reward values, ensuring that the scheduling strategy balances multiple objectives and avoids pursuing a single metric while neglecting other critical needs. When the safety risks of transporting specially managed drugs or cold chain drugs increase, the weight of the safety dimension is automatically increased, guiding the strategy to prioritize the transport of high-risk drugs and highlighting the core position of safe delivery. Simultaneously, combined with the grid adaptability feedback from the entropy-driven grid partitioning algorithm, the algorithm dynamically adjusts the allocation of transport capacity in each region, allowing transport resources to be tilted towards areas with high order density and urgent demand, reducing resource idleness and improving capacity utilization. After the generated scheduling strategy is synchronized to the big data resource service platform, it provides clear and scientific guidance for actual delivery execution, ensuring that every order dispatch and every delivery route is optimally calculated, improving overall delivery efficiency and service quality. Figure 3 As shown.

[0026] A real-time feedback closed-loop iteration mechanism is deployed to build a dynamic optimization system covering the entire process. Comprehensive multi-dimensional data on actual delivery results is collected to ensure that optimization is based on solid evidence. Delivery cost data is automatically generated by the platform from order settlement data and resource scheduling logs, including unit order delivery cost and delivery resource idle rate. By analyzing this data, weak links in cost control can be accurately identified, and resource allocation plans can be adjusted accordingly to reduce unnecessary expenses. Delivery timeliness data is uploaded in real-time via GPS positioning of delivery terminals, automatically comparing the expected delivery time with the actual delivery time to calculate the deviation and generate data on actual delivery time and the percentage of late orders. Based on this data, delivery route planning and capacity allocation can be optimized to reduce lateness. In the pharmaceutical safety data section, cold chain temperature compliance records are automatically uploaded by the cold chain transportation equipment throughout the entire process. Control data generation includes: drug integrity rate confirmed by photos of drug condition taken by delivery personnel's terminals; and special drug escort trajectories recorded by electronic control tags. This data comprehensively monitors drug transportation safety, promptly identifies potential safety hazards, and adjusts control measures accordingly. Customer satisfaction data is generated from user-submitted reviews via the app and customer service-synchronized complaint records, directly reflecting users' genuine feedback and providing direction for service optimization. Grid adaptation data is generated in real-time by an entropy-driven grid partitioning algorithm, providing feedback on the dynamic matching of orders and resources within each grid, including order completion rates and resource matching saturation, offering direct evidence for grid partitioning optimization. Based on this data, the indicator deviation rate is calculated, relevant weights are dynamically adjusted, and the grid partitioning results are iteratively optimized using the entropy-driven grid partitioning algorithm and the entropy-driven reward algorithm. The scheduling strategy is continuously updated, enabling the entire delivery system to quickly respond to changes in order distribution, traffic conditions, drug demand, and user feedback. This achieves a complete closed loop from data collection and decision execution to result feedback and strategy optimization, ensuring continuous improvement in delivery service quality and that resource allocation and scheduling are always optimal.

[0027] This embodiment addresses the complex delivery needs of large-scale pharmaceutical e-commerce platforms in first-tier cities, constructing a complete delivery system through a five-step core process. A data foundation breaks down data barriers to provide precise support; dynamic grid partitioning and a three-level model achieve efficient full-domain management; an adaptation matrix ensures safe and compliant drug delivery; a scheduling strategy balances multi-dimensional objectives; and a closed-loop iterative mechanism drives continuous optimization. The deep application of entropy-driven grid partitioning and entropy-driven reward algorithms, combined with reinforcement learning and adaptation constraints, allows delivery resources to accurately match demand, effectively addressing challenges such as high order density and complex traffic in cities, significantly improving delivery efficiency, safety, and customer satisfaction, and building a stable and efficient full-domain delivery service system.

[0028] Example 2: Example of an online delivery system for small and medium-sized city chain pharmacies This embodiment applies to an online delivery system for chain pharmacies in small and medium-sized cities. The system covers urban built-up areas and surrounding towns, with relatively concentrated store distribution. It needs to meet the needs of local residents for immediate delivery of common medications, standardized delivery of special medications, and safe delivery of cold chain medications. Leveraging the advantages of store resources, through scientific resource allocation and scheduling, it creates an efficient delivery service tailored to local needs. Figure 2 As shown.

[0029] A big data resource service platform for pharmaceuticals has been established, focusing on the business scenarios and local delivery characteristics of chain pharmacies. It comprehensively integrates relevant data resources from each store, forming a core data hub supporting the entire delivery process. Order data is collected from pharmacy online transaction records, including delivery addresses, order quantities, and order times within each store's coverage area. By integrating this data, the platform can accurately grasp the demand intensity and time distribution patterns in each store's catchment area, providing data support for resource allocation and order distribution among stores. Drug attribute data is taken from each store's drug files, clearly defining the category, storage requirements, and regulatory level of each type of drug. This ensures that subsequent delivery methods are accurately matched to drug characteristics, guaranteeing the safety and compliance of drug transportation. Traffic condition data is obtained from local real-time traffic feedback channels and road network monitoring data, covering the traffic conditions and real-time road conditions of urban main roads and rural roads, tailored to small and medium-sized cities. The city's relatively low traffic volume but complex rural road conditions provide a reference for delivery route planning tailored to local conditions. Delivery resource data comes from delivery records at each store, including the status of store delivery personnel, partner delivery vehicles, and the availability of cold chain equipment. This provides a clear understanding of each store's capacity reserves, facilitating coordinated allocation among stores and preventing insufficient capacity or idle resources at any single store. Customer evaluation data is collected from user review submissions and store after-sales feedback, quickly capturing local users' service needs and dissatisfaction, providing direct evidence for optimizing service details. Actual delivery result data is obtained from delivery completion feedback, including delivery completion time and delivery status, providing foundational data for closed-loop optimization. The collected data undergoes initial screening to remove invalid data, ensuring the accuracy and effectiveness of the dataset. Based on this, a unified dataset is constructed, and a real-time data synchronization mechanism is established. This allows for on-demand data retrieval from each store and each stage, breaking the previous isolation of store data. This enables scheduling decisions to comprehensively consider all store resources and overall needs, laying a solid data foundation for the efficient advancement of subsequent stages.

[0030] Based on the geographical boundary data of the city's built-up area and surrounding towns provided by the big data resource service platform, and considering the relatively concentrated delivery range of small and medium-sized cities and the dispersed demand in rural areas, an initial delivery area grid is generated using an equidistant division method. Each grid is assigned a unique number and bound to corresponding geographical latitude and longitude coordinates, enabling precise identification and management of the delivery area. The initial grid is validated by combining historical order distribution data from pharmacies with local traffic network data, eliminating invalid grids in remote areas with no delivery demand. This avoids allocating resources to areas without demand, reducing operating costs. The validated grid information is stored in the platform's grid model library, forming the initial area division template. An entropy-driven grid division algorithm is employed, dynamically adjusting the grid size based on real-time order density, local traffic congestion, and drug demand data. This allows the grid division to accurately adapt to the spatiotemporal changes in local demand; for example, the grid is appropriately enlarged when orders are dispersed in rural areas and reduced when orders are concentrated in urban areas, improving the alignment between the grid and actual demand. Based on dynamically adjusted grids, a three-tiered delivery area model is generated. The city-level delivery area covers the entire urban built-up area and surrounding towns, handling resource coordination and scheduling across towns and districts, balancing delivery capacity between urban areas and towns, and between different stores, avoiding regional capacity imbalances. The district-level delivery area is divided based on urban business districts, town centers, and major transportation routes, covering the core service areas of each store, handling delivery task allocation and capacity deployment within the district, catering to the concentrated nature of stores and improving task allocation efficiency and response speed. The grid-level delivery area, composed of dynamically adjusted grids, accurately covers last-mile delivery needs in communities, towns, and villages, handling last-mile delivery execution and demand response, shortening last-mile delivery distances and improving delivery timeliness in towns. The generated three-tiered delivery area model is synchronized to a big data resource service platform, providing a regional foundation tailored to the characteristics of small and medium-sized city delivery for subsequent adaptation matrix construction and scheduling strategy training, enabling the delivery system to balance efficient urban delivery with precise coverage of towns.

[0031] Based on data from a big data resource service platform and combined with the delivery capabilities and local needs of small and medium-sized cities, a refined classification of drug types and delivery methods is conducted. According to drug storage conditions, regulatory management levels, and transportation risk levels, drugs sold online by pharmacies are divided into three categories: cold chain drugs, specially managed drugs, and ordinary room-temperature drugs. Cold chain drugs include vaccines and biological agents, while specially managed drugs include narcotic drugs and psychotropic drugs, clearly defining the transportation requirements for different drugs. Based on local delivery capabilities and timeliness requirements, delivery methods are divided into five categories: instant delivery, same-day delivery, next-day delivery, dedicated cold chain delivery, and dedicated personnel delivery for special drugs, catering to the daily needs of residents in small and medium-sized cities for instant delivery and the standardized transportation requirements for special drugs. Establish scientific adaptation rules: cold chain medicines are adapted to dedicated cold chain delivery to ensure that storage temperature requirements are met throughout the process. Temperature control data is uploaded in real time during delivery to ensure the efficacy and safety of cold chain medicines. Specially managed medicines are adapted to delivery by designated personnel, escorted by certified personnel throughout the process, with closed-loop management, recording the escort route and medicine status, strictly following regulatory standards, and eliminating transportation risks. Ordinary room temperature medicines are adapted to immediate delivery, half-day delivery, or next-day delivery based on order timeliness requirements. Immediate delivery prioritizes matching nearby delivery resources around the store to shorten delivery time and improve user experience. Half-day and next-day delivery coordinates transit resources and trunk line delivery resources within the area to balance timeliness and cost. By combining the characteristics of drug demand and the distribution of delivery resources in each grid with the feedback from the entropy-driven grid partitioning algorithm, the dynamic iteration of the adaptation matrix is ​​completed. This allows the adaptation rules to fit the demand structure and resource reserves of different regions. For example, rural areas focus on half-day and next-day delivery, while urban areas focus on instant delivery. After being stored in the big data resource service platform, it provides a clear matching basis for the scheduling strategy, ensuring that all types of drugs can obtain suitable delivery solutions. This ensures safety and compliance while meeting the service expectations of local residents.

[0032] Based on a three-tiered delivery area model, and considering the characteristics of concentrated stores and relatively limited delivery capacity in small and medium-sized cities, a reinforcement learning state space and action space are constructed. The state space includes the reserve of core resources such as cold chain equipment and dedicated delivery personnel in city-level delivery areas; the real-time order density in district-level delivery areas; and the distribution of store delivery personnel and cooperative delivery capacity in grid-level delivery areas. This comprehensive understanding of the core status of the delivery system provides a complete reference for scheduling decisions. The action space covers order dispatch rules, cross-regional resource allocation paths, and last-mile delivery priority settings, aligning with the needs of coordinated allocation between stores. By incorporating the constraints of a drug-delivery method adaptation matrix, the matching range of delivery resources for various drugs is clearly defined, preventing incompatible deliveries. An entropy-driven dynamic reward algorithm is employed to comprehensively evaluate delivery costs, timeliness, drug safety, and customer satisfaction. Dynamic reward values ​​optimize scheduling actions, balancing multi-dimensional objectives and avoiding overall service imbalance caused by optimizing a single indicator. When the risk of temperature fluctuations increases during cold chain drug transportation or when the control requirements for the transportation of special drugs are strengthened, the weight of the safety dimension is automatically increased to prioritize the transportation safety of high-risk drugs. Simultaneously, based on the grid adaptability feedback from the entropy-driven grid partitioning algorithm, the allocation of transportation capacity in each area and grid is dynamically adjusted. This ensures that transportation resources are tilted towards urban grids with concentrated orders and rural grids with dispersed demand but high response difficulty, guaranteeing that store transportation capacity can accurately match order demand and avoiding resource waste and capacity gaps. The generated scheduling strategy, after being synchronized to the big data resource service platform, provides clear guidance for actual delivery execution, achieving efficient collaboration between stores and regions, improving overall delivery efficiency and service quality, and aligning with the current delivery resource situation in small and medium-sized cities.

[0033] Deploy a real-time feedback closed-loop iteration mechanism, and combine it with the characteristics of delivery scenarios in small and medium-sized cities to comprehensively collect multi-dimensional actual delivery result data, providing accurate basis for system optimization. Delivery cost data is automatically generated by the platform from order settlement data and resource scheduling logs of each store, producing unit order delivery cost and delivery resource idle rate. By analyzing this data, resource allocation between stores can be optimized, reducing overall operating costs. Delivery timeliness data is generated by uploading delivery time in real time through GPS positioning of delivery terminals. The deviation is calculated by comparing with the expected delivery time, forming the actual delivery time and the proportion of overdue orders. Based on this data, delivery routes and capacity allocation can be optimized to reduce overdue situations, especially improving delivery timeliness in rural areas. In terms of drug safety data, cold chain temperature compliance is automatically confirmed by data uploaded by cold chain equipment, drug integrity rate is verified by photos taken by delivery personnel terminals, and the escort trajectory of special drugs is recorded through electronic control tags, comprehensively monitoring drug transportation safety and promptly identifying and rectifying safety hazards. Customer satisfaction data is generated by integrating user-submitted evaluation information and store after-sales complaint records, quickly responding to local user feedback and optimizing service details. Grid adaptation data is fed back in real time by the entropy-driven grid division algorithm, including the order completion rate and resource matching saturation of each grid, providing a direct basis for grid division and capacity adjustment. Based on these data, the deviation rate of the indicators is calculated, the relevant weights are dynamically adjusted, and the grid partitioning results are iteratively optimized by the co-entropy dynamic grid partitioning algorithm and the entropy-driven state reward algorithm. The scheduling strategy is continuously updated so that the delivery system can quickly adapt to local order fluctuations, changes in traffic conditions and upgrades in user needs. This achieves a complete closed loop from data collection and decision execution to result feedback and strategy optimization, ensuring that the delivery service can continuously meet the business characteristics and residents' needs of small and medium-sized cities, and improving the adaptability and stability of the system.

[0034] This embodiment leverages the characteristics of concentrated chain pharmacy stores in small and medium-sized cities to construct a delivery system tailored to local needs. A data foundation integrates scattered store data, achieving comprehensive resource coordination; dynamic grid partitioning adapts to urban-rural delivery differences, balancing efficiency and coverage; an adaptation matrix aligns with drug characteristics and local delivery capabilities, ensuring safety and compliance; a scheduling strategy balances multiple objectives and optimizes inter-store capacity coordination; and a closed-loop iterative mechanism continuously responds to changes. Through two core algorithms and a multi-level system design, it effectively addresses issues such as urban-rural demand differences and limited delivery capacity, improving delivery timeliness and service quality while controlling costs, providing a feasible solution for online delivery for chain pharmacies in small and medium-sized cities.

[0035] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some modifications or alterations to the above-disclosed technical content to create equivalent embodiments without departing from the scope of the present invention. Any simple modifications, equivalent changes and alterations made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the scope of the present invention.

Claims

1. A method for optimizing the allocation and scheduling of multimodal delivery resources on an online drug platform, characterized in that, The specific steps of this method are as follows: S100, Data Foundation Construction: Build an online big data resource service platform for pharmaceuticals, collect multi-source heterogeneous data, construct a unified dataset after data preprocessing, store the data and complete real-time data synchronization and on-demand retrieval; S200, Dynamic Grid Partitioning: An entropy-driven grid partitioning algorithm is used to construct an initial delivery area grid based on real-time data from a big data resource service platform. The grid size is dynamically adjusted, and a three-level delivery area model is generated and synchronized to the big data resource service platform. The mathematical expression of the entropy-driven grid partitioning algorithm is: in, Divide the dynamic grid into entropy coefficients; The order density is the spatiotemporal entropy; The spatiotemporal entropy of traffic congestion; The spatiotemporal entropy of drug demand; For time The changing attenuation factor; , , The weights are dynamic. The three-level delivery area model is a multi-level delivery area system built on a dynamically adjusted grid. The structure includes three levels: the first level is the city-level delivery area, which is divided according to the city's administrative geographical boundaries, covering the entire delivery geographical area and carrying the functions of cross-regional scheduling and resource coordination; the second level is the area-level delivery area, which is divided according to the city's business districts and main traffic arteries, covering the main delivery concentration areas within the city and carrying the functions of delivery task allocation and capacity deployment within the area; the third level is the grid-level delivery area, which is composed of dynamically adjusted grids, covering the last-mile delivery demand points and carrying the functions of last-mile delivery execution and demand response. The generated three-level delivery area model is synchronized to the big data resource service platform. S300, Adaptation Matrix Construction: Based on the data of the big data resource service platform, drug types and delivery method types are classified, adaptation rules are established, a drug-delivery method adaptation matrix is ​​constructed, and the matrix is ​​dynamically iterated by combining the entropy dynamic grid partitioning algorithm with the feedback data, and stored in the big data resource service platform; S400, Scheduling Strategy Training: An entropy-driven reward algorithm is employed. Based on a three-level delivery area model, a reinforcement learning state space and action space are constructed. Combined with the drug-delivery method adaptation matrix constraint, a scheduling strategy is generated and synchronized to the big data resource service platform. The mathematical expression of the entropy-driven reward algorithm is: in, This is a dynamic reward value; , , , Dynamic weights; As a reward for delivery costs; Incentives for timely delivery; Rewards for drug safety; Rewards for customer satisfaction; Assign grid entropy reward weights; Divide the dynamic grid into entropy coefficients; S500, closed-loop iterative optimization: Deploy a real-time feedback closed-loop iterative mechanism, collect actual delivery result data to calculate the indicator deviation rate, dynamically adjust relevant weights, coordinate the entropy-driven grid partitioning algorithm and the entropy-driven reward algorithm to iteratively optimize the grid partitioning results, and update the scheduling strategy.

2. The method for optimizing the allocation and scheduling of multimodal delivery resources on an online drug platform according to claim 1, characterized in that, In step S100, during the data foundation construction, multi-source heterogeneous data includes order data, drug attribute data, traffic condition data, delivery resource data, customer evaluation data, and actual delivery result data. Specifically, order data includes delivery address, order quantity, and order time; drug attribute data includes drug category and storage requirements; traffic condition data includes road conditions and real-time traffic information; delivery resource data includes delivery personnel and delivery equipment status; customer evaluation data includes evaluation content and evaluation level; and actual delivery result data includes delivery completion time and delivery status. The data is collected from corresponding record channels: order data from transaction records, drug attribute data from drug archives, traffic condition data from real-time traffic feedback channels, delivery resource data from delivery-related records, customer evaluation data from evaluation submission records, and actual delivery result data from delivery completion feedback. During the collection process, each type of data undergoes preliminary screening to remove obviously invalid data.

3. The method for optimizing the allocation and scheduling of multimodal delivery resources on an online drug platform according to claim 1, characterized in that, In S200, during the dynamic grid partitioning, the initial delivery area grid is a fixed-size square grid covering the entire target delivery service area. The initial size is set to 1000 meters × 1000 meters. Each grid is assigned a unique grid number and bound to corresponding geographical latitude and longitude coordinates. The initial delivery area grid is constructed as follows: based on the geographical boundary data of the target delivery service area provided by the big data resource service platform, an initial grid for the entire area is generated using an equidistant partitioning method. Each grid is assigned a unique grid number and bound to corresponding geographical latitude and longitude coordinates. Then, the initial grid is verified by combining the platform's historical order distribution data and traffic network data. Invalid grids without delivery needs are removed. The size, number, and coordinate information of the verified initial grid are stored in the grid model library of the big data resource service platform.

4. The method for optimizing the allocation and scheduling of multimodal delivery resources on an online drug platform according to claim 1, characterized in that, In the S300 adaptation matrix construction, the main content of classifying drug types and delivery methods is as follows: Based on drug storage conditions, regulatory management levels, and transportation risk levels, drugs are divided into three categories: cold chain drugs, specially managed drugs, and ordinary room temperature drugs. Cold chain drugs include biological agents and vaccines, which require low-temperature storage throughout the entire process and real-time temperature control during transportation. Specially managed drugs include narcotic drugs and psychotropic drugs, which must comply with special control regulations. Ordinary room temperature drugs include oral and topical preparations that do not require special storage and transportation conditions. When classifying delivery methods, based on delivery time requirements, transportation conditions, and safety control levels, delivery methods are divided into five categories: instant delivery, half-day delivery, next-day delivery, dedicated cold chain delivery, and dedicated personnel delivery for special drugs. Instant delivery corresponds to delivery needs within 30 minutes, half-day delivery corresponds to delivery needs within 4 hours, next-day delivery corresponds to delivery needs within 24 hours, dedicated cold chain delivery corresponds to delivery needs requiring low-temperature control throughout the entire process, and dedicated personnel delivery for special drugs corresponds to delivery needs requiring closed-loop management by dedicated personnel. The results of these two classifications serve as the basic input for establishing adaptation rules.

5. The method for optimizing the allocation and scheduling of multimodal delivery resources on an online drug platform according to claim 1, characterized in that, In the S300 adaptation matrix construction, the adaptation rules are based on the needs of the drug type and combined with the functional characteristics of the delivery method. The specific content of the adaptation rules is as follows: Cold chain drugs must be adapted to dedicated cold chain delivery, and must meet the low temperature control requirements of 2-8℃ throughout the process. Temperature control data must be uploaded in real time during the delivery process; Specially managed drugs must be adapted to dedicated special drug delivery, and must be escorted by certified personnel throughout the process. The delivery process is subject to closed-loop management, and the trajectory of the escort personnel and the status of the drugs are recorded throughout the process. Transferring or storing drugs in violation of regulations is strictly prohibited; Ordinary room temperature drugs can be adapted to immediate delivery, half-day delivery, and next-day delivery. The adaptation is determined based on the order's delivery time requirements. Immediate delivery should prioritize matching nearby delivery resources, half-day delivery should coordinate regional transit resources, and next-day delivery should be reasonably planned in conjunction with trunk line delivery resources; The adaptation rules explicitly prohibit unsuitable combinations across types, and supplementary constraints are made in conjunction with the drug transportation risk level. Drugs with high risk levels need to have higher delivery control standards.

6. The method for optimizing the allocation and scheduling of multimodal delivery resources on an online drug platform according to claim 1, characterized in that, In the S400 scheduling strategy training, the specific content of the scheduling strategy is as follows: Based on the state of each level of the three-level delivery area model, a reinforcement learning state space is constructed, including the resource reserve of the city-level delivery area, the order density of the urban area-level delivery zone, and the capacity distribution of the grid-level delivery area; an action space is constructed, including order dispatch rules, cross-regional resource allocation paths, and last-mile delivery priority settings. Combined with the constraints of the drug-delivery method adaptation matrix, it is determined that cold chain drugs are only matched with cold chain-specific delivery resources, specially managed drugs are only matched with dedicated personnel delivery resources, and ordinary room temperature drugs are matched with time-appropriate delivery methods as needed. An entropy-driven dynamic reward algorithm is adopted, and the scheduling action is evaluated through the dynamic reward value D-MOR. When the drug safety risk increases, the weight of the safety dimension is automatically increased to guide the strategy to prioritize the transportation of high-risk drugs; combined with the grid adaptation degree fed back by the entropy-driven grid partitioning algorithm, the capacity allocation within the region is dynamically adjusted.

7. The method for optimizing the allocation and scheduling of multimodal delivery resources on an online drug platform according to claim 1, characterized in that, In the S500 closed-loop iterative optimization, the actual delivery result data includes multi-dimensional core operational data. The specific content and collection methods are as follows: Delivery cost data includes unit order delivery cost and delivery resource idle rate, automatically generated by the big data resource service platform from order settlement data and resource scheduling logs; Delivery timeliness data includes actual delivery time and the proportion of overdue orders, transmitted in real-time by GPS positioning of the delivery terminal, automatically compared with the expected delivery time to calculate the deviation; Drug safety data includes cold chain temperature compliance records, drug integrity rate, and special drug escort trajectory, automatically uploaded by cold chain transportation equipment with full-process temperature control data, and delivery personnel taking photos of the intact condition of drugs through terminals, with special drugs relying on electronic control tags to record the closed-loop escort trajectory; Customer satisfaction data includes user ratings and complaint rates, generated by user-submitted evaluation information from the user-end APP and synchronized complaint records from customer service; Grid adaptation data includes order completion rate and resource matching saturation for each grid, dynamically matched with resources within the grid in real-time by the entropy-driven grid partitioning algorithm.

Images