Real-time retail intelligent distribution scheduling method based on space-time clustering, medium and system

Abstract

The invention discloses a time-space clustering-based instant retail intelligent distribution scheduling method, medium and system, and the method comprises the steps: obtaining the geographic information of a hierarchical geographic unit through an electronic fence, collecting the data of a to-be-distributed order, and calculating a distribution time window through a time window prediction model, so as to generate time-space clustering information; according to the method, orders with overlapped delivery time windows in the same geographic unit are aggregated into task packages with the same address, the task packages are dynamically allocated through a matching algorithm based on the real-time state of a rider, and algorithm parameters are updated according to delivery feedback. According to the invention, the accuracy of order aggregation and the rationality of task allocation are obviously improved, and the refined scheduling of transport capacity resources and the optimization and improvement of the overall distribution efficiency are realized.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of intelligent logistics distribution, in particular to an instant retail intelligent distribution scheduling method, medium and system based on space-time clustering. BACKGROUND

[0002] Instant retail distribution is an important link in the field of e-commerce, and its core goal is to complete efficient delivery of goods within the time expected by users. Currently, the distribution scheduling methods commonly used in the industry mainly rely on nearby order taking by riders or order allocation mode based on simple geographic zoning, lacking effective use of the space-time distribution characteristics of orders. Due to the randomness and suddenness of order generation, especially during peak hours, a large number of scattered orders are prone to cause multiple delivery personnel to repeatedly travel in the same area, not only causing waste of transportation resources, but also affecting the overall distribution efficiency and timeliness. The existing technology is difficult to effectively integrate orders in similar geographic areas within a short time, and lacks a mechanism for dynamic task allocation based on real-time distribution status, resulting in system response lag, low resource utilization, and difficulty in adapting to the increasing demand for distribution agility and economy in the instant retail scenario. Therefore, there is an urgent need for a new distribution method that can comprehensively optimize order aggregation and dynamic scheduling capabilities to improve overall operational efficiency and improve user experience. SUMMARY

[0003] In view of the above problems, the present application provides an instant retail intelligent distribution scheduling method, medium and system based on space-time clustering, which generates same address task packages by order clustering of hierarchical geographic units and dynamic time windows and intelligently matches riders, solving the problems of transportation waste and low distribution efficiency caused by scattered orders during peak hours.

[0004] To achieve the above purpose, in a first aspect, the present application provides an instant retail intelligent distribution scheduling method based on space-time clustering, comprising:

[0005] Obtain the geographic information of the hierarchical geographic units in the target area based on the electronic fence, the geographic information including cell information, building group and building level, and real-time collect the to-be-delivered order data in each hierarchical geographic unit;

[0006] Obtain historical distribution data, and calculate the delivery time window of each order information based on the order information in the historical distribution data and the to-be-delivered order data through a time window prediction model, to obtain space-time clustering information, the order information including purchase commodity information and user expected delivery time;

[0007] Generate same address task packages based on hierarchical geographic units and space-time clustering information, the same address task package being configured to include multiple orders with overlapping delivery time windows in the area where the same geographic unit is located;

[0008] Real-time acquisition of multiple rider information, each rider information including position information and task state;

[0009] The same address task package is dynamically allocated to the rider meeting the preset distribution condition for distribution through a matching algorithm, and the matching algorithm is configured to comprehensively consider the current position of the rider, the estimated arrival time and the task package distribution path;

[0010] And, obtain the distribution feedback information, map and store the distribution feedback information with the rider information, and update the configuration parameters of the matching algorithm according to the distribution feedback information.

[0011] In some embodiments, according to historical distribution data and order information in to-be-delivered order data, the delivery time window of each order information is calculated by a time window prediction model to obtain spatiotemporal clustering information, including:

[0012] Based on historical distribution data, a distribution duration feature library is constructed in units of small areas, and the distribution duration feature library contains path time consumption from stores to various hierarchical geographic units and building group distribution time consumption in different time periods;

[0013] Extract the product feature vector and user feature vector in the to-be-delivered order data, the product feature vector includes product category combination and preservation requirement, and the user feature vector includes historical order time preference and address feature;

[0014] The distribution duration feature library, the product feature vector and the user feature vector are input into the time window prediction model, and the best departure time window and the estimated arrival time interval of each order are calculated by a multi-objective optimization algorithm, and the best departure time window and the estimated arrival time interval are arranged into a delivery time window;

[0015] The order set in the same hierarchical geographic unit and with a delivery time window whose time window overlap degree exceeds a preset overlap threshold is spatiotemporally marked to generate a clustered order group with spatiotemporal correlation, denoted as spatiotemporal clustering information.

[0016] In some embodiments, the distribution duration feature library, the product feature vector and the user feature vector are input into the time window prediction model, and the best departure time window and the estimated arrival time interval of each order are calculated by a multi-objective optimization algorithm, and the best departure time window and the estimated arrival time interval are arranged into a delivery time window, including:

[0017] Based on the path time consumption, the real-time distribution reference duration of each hierarchical geographic unit under the current traffic condition is calculated by a dynamic path planning algorithm;

[0018] According to the preservation requirement parameter in the product feature vector and the time preference parameter in the user feature vector, a constraint satisfaction algorithm is used to individually adjust the real-time distribution reference duration to generate an initial time window set;

[0019] The initial time window set is processed by using a multi-objective optimization algorithm, which simultaneously optimizes the delivery efficiency target, the product freshness target, and the user satisfaction target, and outputs the optimized time window parameters of each order;

[0020] The upper limit of the checkout time and the lower limit of the delivery time in the optimized time window parameters are combined and matched to generate the best checkout time window and the expected delivery time interval with a flexible interval, and the flexible interval includes a time buffer threshold that allows the rider task allocation;

[0021] The best checkout time window and the expected delivery time interval are arranged into a delivery time window;

[0022] In addition, the generated delivery time window is subjected to conflict detection, and when an overlap conflict is detected in the delivery time window of the same rider, a time window adjustment algorithm is used to reassign the priority of the delivery time window.

[0023] In some embodiments, the same address task package is generated based on hierarchical geographic units and spatio-temporal clustering information, including:

[0024] According to the building group and the building level, a spatial proximity evaluation model is established to calculate the geographical spatial aggregation degree between orders;

[0025] According to the geographical spatial aggregation degree and the overlap degree of the delivery time window, an order set that meets the preset aggregation condition is screened through a spatio-temporal coupling algorithm, which is recorded as an aggregated order set;

[0026] The aggregated order set screened out is subjected to commodity compatibility analysis, and an optimal commodity grouping scheme is generated based on the characteristics of commodity categories, storage temperature, and packaging specifications using a combination optimization algorithm;

[0027] According to the carrying capacity parameters of the rider and the type of delivery tool, the optimal commodity grouping scheme is subjected to load balancing adjustment to form a task package unit that meets the actual delivery conditions;

[0028] A unique identification code is assigned to each task package unit, and the orders in the task package unit are sorted according to a delivery path optimization algorithm to generate a same address task package containing delivery order suggestions;

[0029] In addition, the state of the task package unit is monitored in real time, and when a new order or order cancellation occurs, a dynamic reorganization algorithm is used to update the affected same address task package.

[0030] In some embodiments, the same address task package is dynamically allocated to a rider who meets the preset delivery conditions for delivery through a matching algorithm, including:

[0031] A rider multi-dimensional evaluation model is established, and dynamic matching degree indexes of each rider are calculated in real time, and the dynamic matching degree indexes include a path distance of a current position and a pickup point of a same address task package, a remaining delivery time length of a current task, a historical similar task completion efficiency, and a carrier adaptation degree;

[0032] A same address task package-rider matching matrix is constructed according to the dynamic matching degree indexes and a reinforcement learning algorithm, a matching weight value between each same address task package and an available rider is calculated in real time, the matching weight value is updated to the same address task package-rider matching matrix, and the matching weight value is set according to a task urgency, a delivery path coincidence degree and a rider skill rating;

[0033] A distributed task allocation algorithm is used to globally optimize the same address task package-rider matching matrix, and an optimized same address task package-rider matching matrix is obtained;

[0034] A rider satisfying a matching threshold in the same address task package-rider matching matrix is selected, and a same address task package is bound to a delivery task corresponding to the rider, and a matching state corresponding to the same address task package is changed to matching success;

[0035] A dynamic delivery route scheme is generated for each same address task package matched successfully, and the dynamic delivery route scheme includes pickup path planning based on real-time traffic, delivery sequence optimization of multiple orders in the task package, and predicted time node labeling;

[0036] When it is detected that the rider deviates from the predetermined route or delivery delay occurs, a task re-allocation process is automatically triggered, and a standby rider is re-matched through a nearest scheduling algorithm.

[0037] In some embodiments, a rider multi-dimensional evaluation model is established, and dynamic matching degree indexes of each rider are calculated in real time, including:

[0038] Rider dynamic position data is collected in real time through a rider terminal, an optimal path distance of a current position of the rider and a pickup point of a same address task package is calculated in combination with a road network topology, and an actual travel time is predicted based on real-time traffic flow data;

[0039] Delivery task information currently undertaken by each rider is extracted from an order management system, a remaining delivery time length of a current task of the rider is calculated by using a time estimation algorithm in combination with task geographic distribution and a remaining delivery amount;

[0040] Historical delivery records of each rider under similar space-time conditions are analyzed, and historical similar task completion efficiency indexes including a delivery punctuality rate and an abnormal handling efficiency are extracted;

[0041] According to a volume, a weight and special requirements of goods of the same address task package, in combination with capacity parameters of a carrier associated with each rider, a carrier adaptation degree of the carrier and the same address task package is calculated.

[0042] Based on the current delivery scene characteristics, the adaptive algorithm is used to dynamically adjust the weight proportions of the four dimensions of the optimal path distance, the current task remaining delivery time length, the historical similar task completion efficiency index and the vehicle adaptation degree;

[0043] The optimal path distance, the current task remaining delivery time length, the historical similar task completion efficiency index and the vehicle adaptation degree are weighted and calculated according to the weight proportions by the multi-dimensional fusion algorithm, and the dynamic matching degree index of the rider is output.

[0044] In some embodiments, a same address task package-rider matching matrix is constructed according to the dynamic matching degree index combined with the reinforcement learning algorithm, and the matching weight value between each same address task package and the available rider is calculated in real time, and the matching weight value is updated to the same address task package-rider matching matrix, including:

[0045] The same address task package characteristics, the rider state and the environmental variables are coded into a multi-dimensional state vector, and the state vector includes the space-time distribution characteristics of the same address task package, the real-time position and the load state of the rider, the road network passing state and the weather influence factors;

[0046] The matching relationship between the rider and the same address task package is abstracted as a discrete action set, the optimal matching strategy is learned through the deep Q network algorithm, and the matching weight value of each candidate matching pair and the matching strategy are output.

[0047] In some embodiments, the delivery feedback information is obtained, the delivery feedback information is mapped and stored with the rider information, and the configuration parameters of the matching algorithm are updated according to the delivery feedback information, including:

[0048] The delivery feedback information is updated to the historical delivery record of the current rider by using the space-time alignment algorithm, and the rider portrait including the historical delivery performance, the abnormal event record and the ability evaluation is generated;

[0049] And, the feedback features including the path deviation rate, the time efficiency achievement degree, the service score fluctuation and the abnormal processing time efficiency are extracted from the delivery feedback information;

[0050] The weight coefficients, the priority threshold and the path planning parameters in the deep Q network algorithm are optimized according to the feedback features.

[0051] In the second aspect, the present application further provides a computer readable storage medium, which stores computer program instructions, the computer program instructions realize the method of the first aspect when executed by a processor.

[0052] In the third aspect, the present application further provides an instant retail intelligent delivery scheduling system based on space-time clustering, which is suitable for the method of the first aspect.

[0053] Distinguished from the prior art, the technical scheme provides an instant retail intelligent distribution scheduling method, medium and system based on space-time clustering, geographical information of hierarchical geographical units is acquired through an electronic fence, and order data to be distributed is collected, a distribution time window is calculated by using a time window prediction model to generate space-time clustering information, orders in the same geographical unit with overlapping distribution time windows are aggregated into the same address task package according to the space-time clustering information, the task package is dynamically distributed through a matching algorithm based on the real-time state of a rider, and algorithm parameters are updated according to distribution feedback.

[0054] The above content is only a summary of the technical scheme of the present application, in order to enable those skilled in the art to more clearly understand the technical scheme of the present application, and then can be implemented according to the content of the description and the drawings, and in order to make the above purpose and other purposes, characteristics and advantages of the present application more easily understood, the following is described in combination with the specific embodiments and the drawings of the present application. BRIEF DESCRIPTION OF DRAWINGS

[0055] The drawings are only used to show the principles, implementation manners, applications, characteristics and effects of the specific embodiments and other related contents of the present application, and cannot be considered as a limitation of the present application.

[0056] In the drawings of the specification:

[0057] Figure 1 The method step diagram of steps S101 to S105 of the scheduling method described in the specific embodiment;

[0058] Figure 2 The method step diagram of steps S201 to S204 of the scheduling method described in the specific embodiment;

[0059] Figure 3 The method step diagram of steps S301 to S305 of the scheduling method described in the specific embodiment;

[0060] Figure 4 The method step diagram of steps S401 to S405 of the scheduling method described in the specific embodiment;

[0061] Figure 5 The method step diagram of steps S501 to S505 of the scheduling method described in the specific embodiment. DETAILED DESCRIPTION

[0062] In order to explain the possible application scenarios, technical principles, specific implementation schemes, and the purposes and effects of the present application in detail, the following will be described in detail in combination with the specific embodiments listed and the accompanying drawings. The embodiments described herein are only used to more clearly illustrate the technical solutions of the present application, and therefore only serve as examples, but cannot limit the protection scope of the present application.

[0063] In this paper, the term "embodiment" means that the specific features, structures or properties described in combination with the embodiment can be included in at least one embodiment of the present application. The term "embodiment" appearing at various places in the specification does not necessarily refer to the same embodiment, and does not particularly limit its independence or association with other embodiments. In principle, in the present application, as long as there is no technical contradiction or conflict, each technical feature mentioned in each embodiment can be combined in any way to form a corresponding implementable technical solution.

[0064] Unless otherwise defined, the meanings of the technical terms used herein are the same as those commonly understood by those skilled in the art to which the present application belongs; the use of related terms herein is only for the purpose of describing specific embodiments, and is not intended to limit the present application.

[0065] In the description of the present application, the phrase "and / or" is a description of the logical relationship between the objects, which means that there can be three relationships, for example, A and / or B, which means that there are three cases: A exists, B exists, and A and B exist at the same time. In addition, the character " / " in this paper generally represents that the associated objects before and after are a kind of "or" logical relationship.

[0066] In the present application, the phrases such as "first" and "second" are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual quantity, primary and secondary or order relationship between the entities or operations.

[0067] In the present application, without more limitation, the "includes", "contains", "has" or other similar open expressions used in the sentence are intended to cover non-exclusive inclusion, and these expressions do not exclude the presence of other elements in the process, method or product including the described elements, so that the process, method or product including a series of elements can not only include those limited elements, but also include other elements not explicitly listed, or also include the elements inherent to such process, method or product.

[0068] As the same understanding in the "Examination Guidelines", in this application, "greater than", "less than", "exceed" and so on are understood as not including the number; "above", "below", "within" and so on are understood as including the number. In addition, in the description of the embodiments of the application, the meaning of "multiple" is more than two (including two), and similar expressions related to "multiple" are also understood in this way, for example, "multiple groups", "multiple times" and the like, unless otherwise explicitly and specifically limited.

[0069] In the description of the embodiments of the application, the spatially related expressions used, such as "center", "longitudinal", "transverse", "length", "width", "thickness", "up", "down", "front", "back", "left", "right", "vertical", "horizontal", "vertical", "top", "bottom", "inside", "outside", "clockwise", "counterclockwise", "axial", "radial", "circumferential", and the like, indicate the orientation or positional relationship based on the orientation or positional relationship shown in the specific embodiment or the drawing, and are only for the convenience of describing the specific embodiments of the application or for the reader to understand, and do not indicate or imply that the indicated device or component must have a specific position, a specific orientation, or be constructed or operated in a specific orientation, and therefore cannot be understood as a limitation on the embodiments of the application.

[0070] The processor described in the embodiments of the application can be realized by hardware, firmware, software or a combination thereof, and can use at least one of circuit, single or multiple application specific integrated circuits (ASIC), digital signal processors (DSP), digital signal processing devices (DSPD), programmable logic devices (PLD), field programmable gate arrays (FPGA), central processing units (CPU), controllers, microcontrollers, microprocessors, and other physical, biological or chemical structures that can realize the same or equivalent functions as the above-mentioned processors, such as biological neurons, quantum computing units, DNA computing units, etc., so that the processor can execute part or all steps or any combination of steps mentioned in the computer programs or methods of various embodiments of the application.

[0071] Please refer to Figure 1 In the first aspect, the embodiment provides an instant retail intelligent distribution scheduling method based on space-time clustering, comprising:

[0072] S101, obtain geographical information of hierarchical geographic units in a target area based on an electronic fence, the geographical information including cell information, building group and building level, and real-time collection of to-be-delivered order data in each hierarchical geographic unit;

[0073] S102, obtain historical delivery data, and calculate a delivery time window for each order information based on the historical delivery data and order information in the to-be-delivered order data through a time window prediction model, to obtain spatio-temporal clustering information, the order information including purchase commodity information and user expected delivery time;

[0074] S103, generate a same address task package based on the hierarchical geographic unit and the spatio-temporal clustering information, the same address task package being configured to include multiple orders with overlapping delivery time windows in the area where the same geographic unit is located;

[0075] S104, real-time acquisition of a plurality of rider information, each rider information including position information and task state;

[0076] S105, dynamically assigning the same address task package to a rider meeting a preset delivery condition for delivery through a matching algorithm, the matching algorithm being configured to comprehensively consider the current position of the rider, the estimated arrival time and the task package delivery path;

[0077] and obtaining delivery feedback information, mapping and storing the delivery feedback information with the rider information, and updating the configuration parameters of the matching algorithm according to the delivery feedback information.

[0078] In step S101, the electronic fence is a virtual geographical boundary pre-defined by geographic information system (GIS) technology, used to automatically identify and obtain the geographical information of the hierarchical geographic units in the target area. The hierarchical geographic unit is a delivery area unit divided by spatial granularity, wherein the cell information contains the cell name and boundary coordinates, the building group represents a set of buildings with adjacent physical locations, and the building level is refined to specific building number and unit information. The to-be-delivered order data is obtained in real time through the order management system, including user address, commodity list and other key information, for subsequent clustering and scheduling.

[0079] In step S102, the time window prediction model is a prediction algorithm based on machine learning, which uses the time efficiency records in the historical delivery data and the order information of the to-be-delivered orders for training and reasoning. The historical delivery data includes the delivery path, actual time consumption and other information of past orders; the purchase commodity information in the order information involves commodity type, storage condition, etc., and the user expected delivery time represents the user's specified time requirement. The model calculates a flexible interval delivery time window for each order through analysis of the above data, and further obtains spatio-temporal clustering information reflecting the spatio-temporal distribution characteristics of the orders.

[0080] In step S103, the same address task package is an order set unit created to improve the efficiency of batch delivery, and its generation depends on the time coupling of spatial division of hierarchical geographic units and spatio-temporal clustering information. Specifically, the system selects orders in the same geographic unit (such as the same building group) and overlaps the delivery time window to aggregate and package. This process is achieved by calculating the overlap degree of the time window and comparing it with the preset overlap threshold, ensuring that the orders in the package can be completed by the same rider in the same period, thereby reducing repeated paths and waiting time.

[0081] In step S104, the rider information is reported in real time through the rider terminal APP, and the location information is provided by the GPS module. The task status includes whether the rider is currently idle, the amount of tasks accepted, the estimated completion time, etc. These information provides real-time decision basis for subsequent dynamic task allocation.

[0082] In step S105, the matching algorithm is a real-time decision model considering multiple factors, which is used to allocate the generated same address task package to appropriate riders. This algorithm considers the current location of the rider, the estimated time to reach the pickup point, the optimized delivery path of multiple orders in the task package, and the current task load of the rider. By calculating the matching degree between the rider and the task package, and dynamically binding the task package to the optimal rider, the precise scheduling of transportation resources and the overall optimization of the delivery path are achieved.

[0083] The delivery feedback information is collected after the order is completed, including actual delivery time, user rating, whether the path deviates, etc. This feedback information is stored in association with the corresponding rider information, which is used to build the performance profile of the rider. In addition, the feedback information is also used to iteratively optimize the configuration parameters of the matching algorithm, such as adjusting the path planning weight or the time estimation coefficient, so that the algorithm has self-learning ability and can continuously adapt to changes in the actual delivery environment.

[0084] In this embodiment, hierarchical geographic information is obtained through electronic fence, and spatio-temporal clustering information and delivery time window of orders are generated by using time window prediction model, and then same address task package is aggregated. Through the matching algorithm, the appropriate rider is dynamically allocated to the task package, and the algorithm parameters are continuously optimized according to the delivery feedback. This method effectively improves the accuracy of order aggregation and rider scheduling, reduces repeated paths and waiting time, and significantly improves the overall delivery efficiency and resource utilization.

[0085] Please refer to Figure 2 In some embodiments, according to the order information in the historical delivery data and the to-be-delivered order data, the delivery time window of each order information is calculated by a time window prediction model to obtain spatio-temporal clustering information, including:

[0086] S201. Construct a delivery time feature library based on historical delivery data, where the library contains the path time from the store to each hierarchical geographical unit and the delivery time within the building group at different time periods.

[0087] S202: Extract product feature vectors and user feature vectors from the order data to be delivered. The product feature vectors include product category combinations and freshness requirements, and the user feature vectors include historical order time preferences and address characteristics.

[0088] S203: Input the delivery time feature library, the product feature vector, and the user feature vector into the time window prediction model, calculate the optimal departure time window and the estimated delivery time interval for each order through a multi-objective optimization algorithm, and organize the optimal departure time window and the estimated delivery time interval into a delivery time window;

[0089] S204: performing spatiotemporal tagging on a set of orders that are in the same hierarchical geographical unit and whose delivery time windows have a time window overlap exceeding a preset overlap threshold, generating a clustered order group with spatiotemporal correlation, which is recorded as spatiotemporal clustering information.

[0090] In step S201, the delivery time feature database is a historical delivery time database built on a community-based basis. By analyzing a large amount of historical delivery data, it compiles and stores the standard route time from each store to a designated hierarchical geographic unit (e.g., community, building group) over different time periods, as well as the typical time required to complete the last few hundred meters within a building group. This feature database provides an important benchmark for subsequent time window predictions.

[0091] The product feature vector in step S202 is a numerical feature extracted from the delivery order to describe the product attributes. It primarily includes product category combinations (e.g., fresh produce, daily necessities) and corresponding preservation requirements (e.g., temperature control level, time urgency). The user feature vector is a set of features that characterize user preferences. It includes historical order time preferences (e.g., noon delivery preference) derived from past order analysis, as well as address characteristics (e.g., building entrance location, whether upstairs delivery is required, etc.). These feature vectors collectively provide input for personalized time window prediction.

[0092] In step S203, the time window prediction model receives as input the delivery duration feature library, product feature vectors, and user feature vectors. The model's core utilizes a multi-objective optimization algorithm that simultaneously balances multiple objectives, including delivery efficiency, product freshness, and user satisfaction, to calculate the optimal departure time window (i.e., the recommended pickup time range for the delivery driver) and the estimated delivery time interval for each order. Ultimately, these two time intervals are consolidated and merged to form a flexible delivery time window, providing a temporal benchmark for order aggregation.

[0093] In step S204, the preset overlap threshold is a configurable system parameter used to determine whether the delivery time windows of two orders are sufficiently close to be clustered. The system calculates the degree of overlap between the delivery time windows of all orders within the same hierarchical geographic unit (e.g., a group of buildings). When the overlap of a group of orders exceeds the preset threshold, they are considered highly correlated in time and space and are marked as a clustered order group, with the spatiotemporal clustering information recorded. This process automatically groups orders in both spatiotemporal and temporal dimensions, laying a solid foundation for generating task packages for the same address.

[0094] This embodiment builds a delivery time feature library and extracts product and user feature vectors, which are then fed into a time window prediction model. Using a multi-objective optimization algorithm, the system calculates the delivery time window for each order. Based on a preset overlap threshold, the system automatically clusters orders with similar temporal and spatial proximity, generating spatiotemporal clustering information. This embodiment achieves precise order aggregation across time and space, providing a core foundation for subsequent packaging and scheduling, effectively improving the efficiency of batch delivery and the rationality of route planning.

[0095] See also Figure 3 In some embodiments, the delivery time feature library, product feature vectors, and user feature vectors are input into a time window prediction model. The optimal departure time window and estimated delivery time interval for each order are calculated using a multi-objective optimization algorithm. The optimal departure time window and estimated delivery time interval are then organized into a delivery time window, including:

[0096] S301, calculating the real-time delivery benchmark duration of each hierarchical geographical unit under the current traffic conditions based on the path consumption through a dynamic path planning algorithm;

[0097] S302: Based on the freshness requirement parameter in the product feature vector and the time preference parameter in the user feature vector, a constraint satisfaction algorithm is used to perform personalized adjustment on the real-time delivery benchmark duration to generate an initial time window set.

[0098] S303. Process the initial time window set using a multi-objective optimization algorithm. The multi-objective optimization algorithm simultaneously optimizes the delivery efficiency goal, the product freshness goal, and the customer satisfaction goal, and outputs the optimized time window parameters for each order.

[0099] S304: Combine and match the upper limit of the departure time and the lower limit of the delivery time in the optimized time window parameters to generate an optimal departure time window and an estimated delivery time interval with a flexible interval. The flexible interval includes a time buffer threshold that allows for rider task allocation.

[0100] S305: Arrange the optimal check-out time window and the estimated delivery time interval into a delivery time window;

[0101] and, the generated delivery time window is subjected to conflict detection, and when an overlap conflict is detected in the delivery time window of the same rider, a time window adjustment algorithm is used to reassign the priority of the delivery time window.

[0102] In step S301, the dynamic path planning algorithm is an algorithm for real-time calculation of the optimal path from the store to each hierarchical geographic unit, which dynamically corrects the historical path time consumption by accessing real-time traffic flow data, road closure events and weather conditions, thereby outputting the real-time delivery reference time length under the current traffic condition, which provides accurate and time-effective reference input for subsequent time window calculation.

[0103] In step S302, the constraint satisfaction algorithm is used to process multiple constraints, which adjusts the real-time delivery reference time length according to the preservation requirement parameters in the commodity feature vector (such as the maximum allowed delivery time length required by fresh commodities) and the time preference parameters in the user feature vector (such as the user-specified expected delivery time period), generates an initial time window set that satisfies various constraints, and ensures that the time window meets both objective delivery conditions and subjective user needs.

[0104] In step S303, the multi-objective optimization algorithm optimizes the delivery efficiency target (such as minimizing the total delivery time), the commodity preservation target (such as ensuring that perishable goods are delivered quickly), and the user satisfaction target (such as meeting the user's expected time as much as possible). The algorithm processes the initial time window set, outputs a set of optimized time window parameters, and achieves the best balance between multiple objectives.

[0105] In step S304, the upper limit of the departure time and the lower limit of the arrival time in the optimized time window parameters are combined and matched to generate the best departure time window and the expected arrival time interval with a flexible interval. This flexible interval includes a time buffer threshold that allows the rider to assign tasks, which is a system preset value, to absorb minor fluctuations in the delivery process, provide reasonable flexibility for the rider, and enhance the robustness of the plan to actual interference.

[0106] Finally, the generated delivery time window is subjected to conflict detection, and when an overlap conflict is detected in the delivery time window of the same rider, a time window adjustment algorithm is used to reassign the priority of the delivery time window. The time window adjustment algorithm usually dynamically adjusts the order of conflicting time windows based on factors such as the urgency of the order, the value of the commodity or the user level, so as to ensure that the task package finally assigned to the rider is feasible and efficient in the time dimension.

[0107] The embodiment calculates the real-time distribution reference duration by a dynamic path planning algorithm, and generates an elastic time window integrating distribution efficiency, product preservation and user preference by using constraint satisfaction and multi-objective optimization algorithm. The scheme introduces a time buffer threshold to enhance robustness, and ensures scheduling feasibility through conflict detection and priority adjustment. The embodiment significantly improves the accuracy, personalization and actual executability of time window prediction, providing core support for efficient rider task allocation.

[0108] Please refer to Figure 4 In some embodiments, the same address task package is generated based on hierarchical geographic units and spatio-temporal clustering information, including:

[0109] S401, according to the building group and the building level, a spatial proximity evaluation model is established, and the geographical spatial aggregation degree between each order is calculated;

[0110] S402, according to the geographical spatial aggregation degree and the overlap degree of the distribution time window, the order set meeting the preset aggregation condition is filtered through the spatio-temporal coupling algorithm, which is recorded as the aggregated order set;

[0111] S403, the aggregated order set filtered out is subjected to commodity compatibility analysis, and a combination optimization algorithm is used to generate an optimal commodity grouping scheme based on commodity categories, storage temperature and packaging specification characteristics;

[0112] S404, according to the carrying capacity parameters of the rider and the distribution tool type, the optimal commodity grouping scheme is subjected to load balancing adjustment to form a task package unit conforming to the actual distribution conditions;

[0113] S405, a unique identification code is allocated to each task package unit, and the orders in the task package unit are sorted according to the distribution path optimization algorithm to generate a same address task package containing distribution order suggestions;

[0114] In addition, the state change of the task package unit is monitored in real time, and when a new order or order cancellation occurs, the affected same address task package is updated by a dynamic reorganization algorithm.

[0115] In step S401, the spatial proximity evaluation model is a calculation model for quantifying the geographical proximity between different orders. It calculates the actual physical distance or path accessibility distance between the order distribution addresses through the geographic coding information of the building group and the building level, and outputs the geographical spatial aggregation degree index, which provides a quantitative basis for the spatial dimension of subsequent order aggregation.

[0116] In step S402, the spatio-temporal coupling algorithm is used to comprehensively evaluate the geographical spatial aggregation degree between orders and the overlap degree of the delivery time window. The orders are filtered by preset aggregation conditions (such as spatial distance below a threshold and time window overlap degree above a threshold), and the order set that meets the spatio-temporal proximity condition is marked as an aggregated order set, so as to realize the preliminary clustering of orders in the spatio-temporal dimension.

[0117] In step S403, the commodity compatibility analysis judges whether the commodities in multiple orders are suitable for combined delivery based on commodity categories, storage temperature, and packaging specifications, etc. The combination optimization algorithm is used to generate an optimal commodity grouping scheme, which aims to improve the number of commodity categories and the quantity of commodities in a single delivery as much as possible under the premise of meeting the commodity storage requirements (such as avoiding odor mixing and temperature zone conflict) and packaging restrictions, and improving the delivery integration efficiency.

[0118] In step S404, the load balancing adjustment is made to the aforementioned commodity grouping scheme according to the carrying capacity parameters of the rider (such as the box volume of an electric vehicle and the maximum load) and the delivery tool type (such as the configuration of a thermal insulation box). This step ensures that the final task package unit formed meets the carrying limit of the actual delivery equipment in terms of volume, weight, and temperature control conditions, avoiding overloading or equipment mismatch problems.

[0119] In step S405, a unique identification code is assigned to each task package unit to realize traceability management, and a delivery path optimization algorithm is used to sort the orders in the package to generate a same-address task package containing delivery sequence suggestions. Preferably, the delivery sequence suggestions aim to achieve the shortest path or the least number of up and down stairs. Finally, a dynamic recombination algorithm is used to monitor the state of the task package in real time, and when there is an increase or decrease in orders, the composition and order of the affected task package are adjusted in time to ensure the real-time effectiveness and executability of the task package. The whole process realizes the complete task package generation and dynamic updating mechanism from order clustering, commodity compatibility judgment, load adaptation to path optimization.

[0120] The embodiment realizes spatio-temporal clustering of orders through spatial proximity evaluation and spatio-temporal coupling algorithm, generates same-address task packages that meet the actual delivery conditions by combining commodity compatibility analysis and load balancing adjustment, ensures delivery efficiency through unique identification code and path optimization, and uses dynamic recombination algorithm to deal with order changes. The embodiment effectively improves the accuracy of order aggregation and the executability of task packages, significantly reduces repeated paths, optimizes the allocation of transport capacity, and reduces delivery costs.

[0121] Please refer to Figure 5 In some embodiments, the same-address task package is dynamically assigned to a rider who meets the preset delivery conditions for delivery through a matching algorithm, including:

[0122] S501, a rider multi-dimensional evaluation model is established, and a dynamic matching degree index of each rider is calculated in real time, and the dynamic matching degree index includes a path distance of a current position and a pickup point of a same address task package, a remaining delivery time length of a current task, a historical same task completion efficiency, and a carrier adaptation degree;

[0123] S502, a same address task package-rider matching matrix is constructed according to the dynamic matching degree index and a reinforcement learning algorithm, a matching weight value between each same address task package and an available rider is calculated in real time, the matching weight value is updated to the same address task package-rider matching matrix, and the matching weight value is set according to a task urgency, a delivery path coincidence degree and a rider skill rating;

[0124] S503, a distributed task allocation algorithm is used to globally optimize the same address task package-rider matching matrix, and an optimized same address task package-rider matching matrix is obtained;

[0125] S504, a rider satisfying a matching threshold in the same address task package-rider matching matrix is selected, and a same address task package is bound to a corresponding delivery task of the rider, and a matching state corresponding to the same address task package is changed to matching success;

[0126] S505, a dynamic delivery route scheme is generated for each same address task package matched successfully, and the dynamic delivery route scheme includes a pickup path planning based on real-time traffic, a delivery sequence optimization of multiple orders in the task package, and a predicted time node marking;

[0127] When it is detected that the rider deviates from the predetermined route or the delivery is delayed, a task re-allocation process is automatically triggered, and a standby rider is re-matched through a nearest scheduling algorithm.

[0128] In step S501, the rider multi-dimensional evaluation model is a calculation model for comprehensively evaluating the delivery capacity of the rider, and information such as the current position of the rider, the remaining time length of the task, the historical efficiency data and the carrier state is collected in real time, and a dynamic matching degree index is generated through weighted calculation. The index quantitatively reflects the adaptation degree of the rider to execute a specific same address task package, and provides an objective basis for task allocation.

[0129] In step S502, the reinforcement learning algorithm is used to construct and continuously optimize the same address task package-rider matching matrix. The algorithm dynamically calculates the matching weight value according to the task urgency (such as fresh food orders first), the delivery path coincidence degree (such as the degree of going the same way), and the rider skill rating (such as familiarity with the area), and updates the real-time calculation result to the matrix, forming a multi-dimensional weighted matching relationship network.

[0130] In step S503, the distributed task allocation algorithm performs optimization calculation on the global matching matrix, which solves the optimal allocation scheme by parallel processing of multiple task packages and matching combinations of riders, effectively avoids local optimal solution, and improves the calculation efficiency and rationality of large-scale task allocation.

[0131] In step S504, qualified riders are selected from the optimized matrix according to a preset matching threshold, and task packages with the same address are bound to their distribution tasks, and the task package state is updated to matching success. The matching threshold can be dynamically adjusted according to the supply and demand of transport capacity, for example, appropriately relaxing the threshold during peak hours to improve the matching success rate.

[0132] In step S505, a dynamic distribution route scheme is generated for each matching successful task package, which integrates real-time traffic prediction, order distribution sequence optimization in the package, and accurate time node calculation, providing full navigation guidance for riders. When the system detects that the rider deviates from the route or delays through GPS positioning, it automatically triggers the task re-allocation process, uses the nearest scheduling algorithm to quickly match replacement candidates from the standby rider pool, and ensures efficient completion of distribution tasks and system robustness. The entire process realizes a complete task allocation closed loop from rider evaluation, intelligent matching, global optimization to dynamic scheduling.

[0133] The embodiment constructs a dynamic matching matrix through a multi-dimensional rider evaluation model and a reinforcement learning algorithm, uses a distributed optimization algorithm to realize intelligent matching of task packages and riders, and generates a dynamic distribution route scheme. When distribution anomalies occur, the system automatically triggers the nearest re-allocation mechanism. The embodiment significantly improves the accuracy of task allocation and distribution efficiency, ensures optimal utilization of transport capacity resources, and guarantees the reliability and robustness of the distribution system through a dynamic fault-tolerant mechanism.

[0134] In some embodiments, a multi-dimensional rider evaluation model is established to calculate the dynamic matching degree index of each rider in real time, including:

[0135] The rider terminal collects real-time dynamic position data of the rider, calculates the optimal path distance between the rider's current position and the pickup point of the task package with the same address based on the road network topology, and predicts the actual travel time based on real-time traffic flow data;

[0136] Task distribution information currently undertaken by each rider is extracted from the order management system, combined with the geographical distribution of tasks and the remaining distribution volume, and a time estimation algorithm is used to calculate the remaining distribution time of the rider's current task;

[0137] Historical distribution records of each rider under similar spatio-temporal conditions are analyzed to extract historical task completion efficiency indicators including distribution punctuality rate and abnormal handling efficiency;

[0138] According to the cargo volume, weight and special requirements of the same address task package, combined with the capacity parameters of the associated carrier of each rider, the carrier and the carrier adaptation degree of the same address task package are calculated;

[0139] Based on the current distribution scene characteristics, an adaptive algorithm is used to dynamically adjust the weight proportion of four dimensions of optimal path distance, current task remaining distribution time length, historical similar task completion efficiency index and carrier adaptation degree;

[0140] Through a multi-dimensional fusion algorithm, the optimal path distance, the current task remaining distribution time length, the historical similar task completion efficiency index and the carrier adaptation degree are weighted and calculated according to the weight proportion, and the dynamic matching degree index of the rider is output.

[0141] In this embodiment, the optimal path distance is the actual reachable distance from the current position of the rider to the task package pickup point, which is calculated based on the real-time road network topology and traffic flow data. This calculation is performed by accessing the real-time traffic API of a third-party map service, combining road grades, congestion coefficients and other parameters for dynamic path planning, so as to accurately reflect the actual travel cost of the rider to reach the pickup point.

[0142] The current task remaining distribution time length is an index obtained by comprehensively analyzing the existing task amount of the rider through a time estimation algorithm. The time estimation algorithm considers the geographical distribution density of the task, the historical data of the average distribution time of a single order, and the current traffic conditions, and predicts the time required by the rider to complete the existing task through a regression model, providing a time dimension basis for evaluating the rider's order acceptance ability.

[0143] The historical similar task completion efficiency index is performance data extracted from the historical distribution records of the rider, mainly including key indicators such as distribution punctuality rate and abnormal order processing time efficiency. Through data mining technology, the historical performance of the rider under similar weather, time period and regional conditions is analyzed to form an objective evaluation of his distribution ability.

[0144] The carrier adaptation degree is a matching degree index obtained by comparing the characteristics of the task package and the parameters of the rider's carrier. According to the total volume, weight and temperature control requirements of the task package, and the box volume, load limit and equipment configuration of the rider's registered carrier, the matching degree is calculated to ensure that the carrier capacity meets the distribution demand.

[0145] The adaptive weight adjustment algorithm dynamically optimizes the weight proportion of each dimension according to the real-time distribution scene characteristics. By monitoring the current supply and demand ratio of transport capacity, weather conditions, regional order density and other environmental factors, the fuzzy logic control principle is used to automatically adjust the contribution of each dimension in the matching degree calculation, so that the evaluation model has environmental adaptability.

[0146] The multi-dimensional fusion algorithm combines the evaluation values of the four dimensions into a final dynamic matching degree index through weighted summation. After the dimensional data is standardized, it is linearly weighted according to the weight proportion determined by the adaptive algorithm, and finally outputs a matching degree score in the range of 0-1, providing a quantitative decision basis for rider selection.

[0147] The embodiment collects multi-dimensional data such as rider position, task load, historical performance and vehicle parameters in real time, dynamically adjusts the contribution proportion of each dimension using an adaptive weight algorithm, and finally outputs the dynamic matching degree index of the rider through a multi-dimensional fusion algorithm, realizing accurate quantitative evaluation of the rider's distribution ability, significantly improving the accuracy and efficiency of task allocation, and ensuring the optimal matching of transportation resources and distribution demand.

[0148] In some embodiments, a same-address task package-rider matching matrix is constructed according to the dynamic matching degree index combined with a reinforcement learning algorithm, and the matching weight value between each same-address task package and available rider is calculated in real time, and the matching weight value is updated to the same-address task package-rider matching matrix, including:

[0149] The same-address task package features, rider state and environmental variables are coded into a multi-dimensional state vector, including the spatio-temporal distribution features of the same-address task package, the real-time position and load state of the rider, the road network traffic state and the weather influencing factors;

[0150] The matching relationship between the rider and the same-address task package is abstracted as a discrete action set, and the optimal matching strategy is learned through a deep Q network algorithm, and the matching weight value of each candidate matching pair and the matching strategy are output.

[0151] In the embodiment, the multi-dimensional state vector is a feature set formed by digitizing and coding the same-address task package features, rider state and environmental variables, specifically including the geographical distribution density of the task package, the time urgency, the real-time position coordinates of the rider, the current number of load tasks, the road congestion index and the weather influencing factors such as temperature and humidity. Heterogeneous data is converted into standardized numerical features through feature engineering methods, providing comprehensive state representation for matching decisions.

[0152] The discrete action set defines the matching relationship between each available rider and the same-address task package as a selectable independent action, generates all possible matching pairs as a candidate action space by arranging and combining the candidate rider list with the task package to be allocated, and converts the matching problem into an action selection problem in reinforcement learning.

[0153] The deep Q-network algorithm refers to a reinforcement learning method using a deep neural network to approximate a Q-value function. The network takes a multi-dimensional state vector as input, calculates the long-term expected return value of each candidate action (matching pair) through multiple layers of nonlinear transformation, i.e., matching weight value, and stabilizes the training process through experience replay and target network mechanisms, finally outputs the optimal matching strategy and corresponding matching weight value. By continuously collecting actual delivery feedback data, taking delivery efficiency improvement and user satisfaction improvement as reward signals, the network parameters are constantly optimized, enabling the matching matrix to have the ability of continuous self-evolution.

[0154] By modeling the delivery matching problem as a reinforcement learning task and dynamically calculating the optimal matching weight of the rider and the task package using the deep Q-network algorithm, this embodiment can adapt to complex and variable delivery environments, continuously optimize the matching strategy, significantly improve the allocation efficiency of task packages with the same address and the overall response speed of the system, and reduce delivery costs and improve user satisfaction through intelligent matching.

[0155] In some embodiments, delivery feedback information is obtained, the delivery feedback information is mapped and stored with rider information, and the configuration parameters of the matching algorithm are updated according to the delivery feedback information, including:

[0156] The delivery feedback information is updated to the historical delivery records of the current rider using a space-time alignment algorithm to generate a rider portrait containing historical delivery performance, abnormal event records, and ability assessment;

[0157] And, feedback features including path deviation rate, time efficiency achievement, service score fluctuation, and abnormal handling efficiency are extracted from the delivery feedback information.

[0158] The weight coefficients, priority threshold values, and path planning parameters in the deep Q-network algorithm are optimized according to the feedback features.

[0159] In this embodiment, the space-time alignment algorithm accurately matches the delivery feedback information with the rider trajectory data based on timestamps and geographic location information. By unifying the time reference and coordinate conversion, the delivery completion time, path trajectory points, user evaluation, and other feedback data are bound with the rider's identity identifier, ensuring the space-time consistency of data mapping.

[0160] The rider portrait refers to a digital file of the rider's ability formed by aggregating historical delivery data, containing historical performance data such as delivery punctuality rate, abnormal order handling efficiency, service score trend, and abnormal event records such as equipment failure and traffic control. Through multi-dimensional data fusion, a comprehensive evaluation system of the rider's delivery ability is formed.

[0161] The feedback features refer to quantitative evaluation indexes extracted from the delivery feedback information, wherein the path deviation rate is calculated by comparing the difference between the actual trajectory and the planned path; the time efficiency achievement degree is determined based on the ratio of the actual delivery time to the estimated time; the service score fluctuation uses the standard deviation of the user score in the sliding window to calculate; and the abnormal handling time efficiency records the time interval from the occurrence of the abnormality to the solution. These features are processed through data cleaning and standardization to form a quantifiable evaluation vector.

[0162] The weight coefficient optimization step dynamically adjusts the contribution weight of each state feature in the deep Q network according to the feedback features. The feature weight gradient is calculated through the back propagation algorithm, and the parameter update is performed with the maximum delivery efficiency and the optimal user satisfaction as the objective function. The priority threshold adjustment sets the matching priority threshold value based on the historical delivery performance distribution, and the path planning parameter is dynamically updated according to the actual path passing efficiency to update the road network passing cost function. Through continuous parameter optimization, the matching algorithm has online learning ability, and continuously improves the overall performance of the delivery system.

[0163] The embodiment realizes fine evaluation of the delivery process by establishing the rider profile and feedback feature extraction mechanism, ensures data accuracy based on the space-time alignment algorithm, and continuously optimizes the configuration parameters of the deep Q network to make the matching algorithm have adaptive evolution ability. The embodiment significantly improves the learning efficiency and matching accuracy of the delivery system, effectively reduces the delivery abnormality rate, and at the same time optimizes the rider delivery experience and user service satisfaction through personalized parameter adjustment.

[0164] In the second aspect, the embodiment also provides a computer readable storage medium, which stores computer program instructions, and the computer program instructions realize the method in the first aspect when executed by a processor.

[0165] The computer program involved in the embodiment can be stored in a computer device readable storage medium, including but not limited to magnetic disk, magnetic tape, magnetic card, floppy disk, flash memory, optical disc, optical card, read-only memory (ROM), random access memory (RAM), erasable programmable ROM (EPROM) and electrically erasable programmable ROM (EEPROM), etc., and also includes other biological, physical or chemical structures that can realize the same or equivalent functions as the above listed storage media, such as DNA, RNA, protein and other units with information storage ability, etc. In specific embodiments, the storage medium involved can be one of the above medium types, or a combination of the above medium types. In different embodiments, the computer program involved in the embodiment can be centrally stored in a single medium, or distributedly stored in multiple media. The storage medium containing the computer device readable storage medium can be a non-volatile memory or a random access memory. These computer device readable storage media can be built-in in the device, or connected with the device as an external device or part of the external device. In some embodiments, the storage medium with the computer device readable storage medium is deployed locally; in other embodiments, the storage medium can also be deployed remotely from the processor, such as network attached storage accessed via RF circuit or external port and communication network, wherein the communication network can be Internet, one or more intranets, local area network (LAN), wide area network (WAN), storage area network (SAN) and the like, or appropriate combination thereof, as long as the access of the computer device to the storage medium can be realized. In addition, the computer program involved in the embodiment can be stored in plaintext / ciphertext form, or can be designed as training data, and integrated and reorganized by model training to be implicitly saved in the parameter state of deep neural network or other machine learning model.

[0166] In a third aspect, the embodiment also provides a real-time retail intelligent distribution scheduling system based on spatio-temporal clustering, which is suitable for the method of the first aspect.

[0167] In the embodiment, the real-time retail intelligent distribution scheduling system based on spatio-temporal clustering is a software and hardware integrated platform integrating order aggregation, path planning and dynamic matching functions, deployed in a cloud server cluster, realizing high concurrency data processing through a distributed computing architecture, including an order access module, a geographic fence engine, a time window calculation unit and an intelligent matching core.

[0168] The order access module is configured to receive and parse retail order data from merchants in real time, extracting key fields such as product information, delivery address, and expected delivery time from the orders. The geofencing engine uses a pre-set hierarchical geographic unit database and a polygon fence algorithm to perform spatial matching on the delivery addresses, achieving precise positioning at the cell-building group-building level. The time window calculation unit dynamically calculates the optimal delivery time window for each order by analyzing historical performance data, combined with traffic conditions and product characteristics.

[0169] The intelligent matching core uses a multi-thread parallel computing architecture to receive rider positioning information and task package status in real time, and generates a rider-task package matching scheme through an operational research optimization algorithm. The system interacts with external map services and weather APIs through a data bus to ensure the real-time and accuracy of dispatching decisions. Optionally, the system supports users initiating a group order invitation, inviting other users within the same geographic unit to place orders together, and when the order volume reaches a pre-set threshold, the system automatically triggers the same address task package generation process.

[0170] The system described in this embodiment performs the method described in the first aspect, significantly reducing repeated path delivery, improving single rider delivery efficiency, reducing delivery costs while improving user satisfaction, and overall achieving efficient use of transportation resources and quality improvement of delivery services.

[0171] Unlike existing technologies, the above technical solutions have the following beneficial effects: hierarchical geographic units and spatio-temporal clustering techniques effectively solve the core problems of scattered transportation capacity and repeated paths in instant retail delivery. Precise geographic positioning is achieved using electronic fences, and elastic delivery windows are generated using a time window prediction model. Intelligent aggregation of orders with the same address forms a task package. Dynamic matching algorithms consider rider location, load status, and real-time traffic conditions to achieve optimal allocation of transportation resources. The system also has self-learning capabilities, adapting to changes in the delivery environment through continuous optimization of algorithm parameters. The above technical solutions significantly improve single rider delivery efficiency and order aggregation, reduce repeated paths and transportation waste, reduce delivery costs while improving user satisfaction, and form an efficient and sustainable intelligent delivery system.

[0172] Finally, it should be noted that although the above embodiments have been described in the specification and drawings of the present application, they do not limit the patent protection scope of the present application. Any equivalent structure or equivalent process substitution or modification based on the essential concept of the present application, using the content described in the specification and drawings, as well as direct or indirect implementation of the technical solutions of the above embodiments in other related technical fields, are all included in the patent protection scope of the present application.

Claims

1. A method for real-time retail intelligent distribution scheduling based on spatiotemporal clustering, characterized by: include: Obtain geographic information of hierarchical geographic units within the target area based on the electronic fence, including cell information, building groups, and building levels, and collect data on pending delivery orders within each hierarchical geographic unit in real time; Obtain historical delivery data and, based on the order information in the historical delivery data and the pending delivery order data, calculate the delivery time window of each order information using a time window prediction model to obtain spatiotemporal clustering information. The order information includes purchased product information and the user's expected delivery time. Generating a same-address task package based on the hierarchical geographic units and the spatiotemporal clustering information, wherein the same-address task package is configured to include multiple orders with overlapping delivery time windows within the area where the same geographic unit is located; Acquire multiple rider information in real time, each rider information including location information and task status; Dynamically assigning the same-address task package to a rider who meets preset delivery conditions for delivery through a matching algorithm, wherein the matching algorithm is configured to comprehensively consider the rider's current location, estimated arrival time, and task package delivery path; Also, obtain delivery feedback information, map and store the delivery feedback information with the rider information, and update the configuration parameters of the matching algorithm based on the delivery feedback information.

2. The instant retail intelligent distribution scheduling method based on spatiotemporal clustering according to claim 1 is characterized in that: Based on the order information in the historical delivery data and the order data to be delivered, the delivery time window of each order information is calculated through the time window prediction model to obtain spatiotemporal clustering information, including: Building a delivery time feature library based on the historical delivery data, with each cell as the unit, wherein the delivery time feature library includes the path time from the store to each hierarchical geographical unit and the delivery time within the building group at different time periods; Extracting a product feature vector and a user feature vector from the to-be-delivered order data, wherein the product feature vector includes product category combinations and freshness requirements, and the user feature vector includes historical order time preferences and address features; Input the delivery time feature library, the product feature vector, and the user feature vector into the time window prediction model, calculate the optimal departure time window and estimated delivery time interval for each order through a multi-objective optimization algorithm, and organize the optimal departure time window and estimated delivery time interval into a delivery time window; A set of orders that are in the same hierarchical geographical unit and whose delivery time windows have a time window overlap exceeding a preset overlap threshold are spatiotemporally marked to generate a clustered order group with spatiotemporal correlation, which is recorded as the spatiotemporal clustering information.

3. The instant retail intelligent distribution scheduling method based on spatiotemporal clustering according to claim 2 is characterized in that: The delivery time feature library, the product feature vector, and the user feature vector are input into the time window prediction model. The optimal departure time window and the estimated delivery time interval for each order are calculated using a multi-objective optimization algorithm. The optimal departure time window and the estimated delivery time interval are then organized into a delivery time window, including: Based on the path time, a dynamic path planning algorithm is used to calculate the real-time delivery benchmark time of each hierarchical geographical unit under the current traffic conditions; Based on the freshness requirement parameter in the product feature vector and the time preference parameter in the user feature vector, a constraint satisfaction algorithm is used to personalize the real-time delivery benchmark duration to generate an initial time window set; Processing the initial time window set using a multi-objective optimization algorithm, wherein the multi-objective optimization algorithm simultaneously optimizes a delivery efficiency goal, a product freshness goal, and a user satisfaction goal, and outputs optimized time window parameters for each order; Combine and match the upper limit of the departure time and the lower limit of the delivery time in the optimized time window parameters to generate an optimal departure time window and an estimated delivery time interval with a flexible interval, wherein the flexible interval includes a time buffer threshold that allows for rider task allocation; Arrange the optimal departure time window and the estimated delivery time interval into a delivery time window; In addition, conflict detection is performed on the generated delivery time windows. When overlapping conflicts are detected in the delivery time windows of the same rider, the time window adjustment algorithm is used to reallocate the priority of the delivery time windows.

4. The instant retail intelligent distribution scheduling method based on spatiotemporal clustering according to claim 1 is characterized in that: Generating a same-address task package based on the hierarchical geographic units and spatiotemporal clustering information includes: Based on building groups and building levels, a spatial proximity evaluation model is established to calculate the geospatial aggregation between orders; According to the geographic spatial aggregation degree and the overlap of the delivery time windows, a set of orders that meet the preset aggregation conditions is screened by a spatiotemporal coupling algorithm and recorded as an aggregated order set; Perform product compatibility analysis on the filtered aggregated order set and use a combinatorial optimization algorithm to generate the optimal product grouping solution based on product category, storage temperature, and packaging specifications. Based on the rider's carrying capacity parameters and the type of delivery tool, the optimal product grouping plan is load-balanced adjusted to form a task package unit that meets actual delivery conditions; Assign a unique identification code to each task package unit, sort the orders in the task package unit according to the delivery path optimization algorithm, and generate a task package with the same address that includes a delivery sequence suggestion; In addition, the status changes of task package units are monitored in real time. When new orders or order cancellations occur, the affected task packages with the same address are updated through a dynamic reorganization algorithm.

5. The instant retail intelligent distribution scheduling method based on spatiotemporal clustering according to claim 1 is characterized in that: The matching algorithm dynamically assigns the same-address task package to a rider who meets the preset delivery conditions for delivery, including: Establish a multi-dimensional evaluation model for riders and calculate the dynamic matching index of each rider in real time. The dynamic matching index includes the path distance between the current location and the pickup point of the task package with the same address, the remaining delivery time of the current task, the historical completion efficiency of similar tasks, and the vehicle adaptability; A same-address task package-rider matching matrix is ​​constructed based on the dynamic matching index combined with a reinforcement learning algorithm. The matching weight value between each same-address task package and the available riders is calculated in real time, and the matching weight value is updated to the same-address task package-rider matching matrix. The matching weight value is set according to the urgency of the task, the overlap of the delivery route, and the rider skill rating; A distributed task allocation algorithm is used to globally optimize the same-address task package-rider matching matrix to obtain the optimized same-address task package-rider matching matrix; Select the rider who meets the matching threshold in the same-address task package-rider matching matrix, bind the same-address task package to the delivery task corresponding to the rider, and change the matching status corresponding to the current same-address task package to matching success; Generate a dynamic delivery route plan for each successfully matched same-address task package. The dynamic delivery route plan includes pickup route planning based on real-time traffic conditions, delivery sequence optimization for multiple orders in the task package, and estimated time node annotation. In addition, when it is detected that the rider deviates from the scheduled route or a delivery delay occurs, the task reallocation process is automatically triggered, and the backup rider is re-matched through the nearest scheduling algorithm.

6. The instant retail intelligent distribution scheduling method based on spatiotemporal clustering according to claim 5 is characterized in that: Establish a multi-dimensional evaluation model for riders and calculate the dynamic matching index of each rider in real time, including: The rider's dynamic location data is collected in real time through the rider's terminal. The optimal path distance between the rider's current location and the pickup point of the task package at the same address is calculated based on the road network topology structure, and the actual travel time is predicted based on real-time traffic flow data. Extract the delivery task information currently undertaken by each rider from the order management system, combine the geographical distribution of the tasks and the remaining delivery volume, and use the time estimation method to calculate the remaining delivery time of the rider's current task; Analyze the historical delivery records of each rider under similar time and space conditions, and extract historical efficiency indicators for similar tasks, including delivery on-time rate and exception handling efficiency; Based on the cargo volume, weight and special requirements of the same-address task package, combined with the capacity parameters of the vehicles associated with each rider, the compatibility of the vehicle with the same-address task package is calculated; Based on the characteristics of the current delivery scenario, an adaptive algorithm is used to dynamically adjust the weight ratios of four dimensions: optimal path distance, remaining delivery time of the current task, historical completion efficiency index of similar tasks, and vehicle adaptability; Through a multi-dimensional fusion algorithm, the optimal path distance, the remaining delivery time of the current task, the historical completion efficiency index of similar tasks and the vehicle adaptability are weighted and calculated according to the weight ratio to output the rider's dynamic matching index.

7. The instant retail intelligent distribution scheduling method based on spatiotemporal clustering according to claim 5 is characterized in that: According to the dynamic matching index combined with the reinforcement learning algorithm, a same-address task package-rider matching matrix is ​​constructed, and the matching weight value between each same-address task package and the available rider is calculated in real time, and the matching weight value is updated to the same-address task package-rider matching matrix, including: Encoding the same-address task package characteristics, rider status, and environmental variables into a multidimensional state vector, wherein the state vector includes the spatiotemporal distribution characteristics of the same-address task package, the rider's real-time location and load status, the road network traffic status, and weather influencing factors; The matching relationship between the rider and the same-address task package is abstracted into a discrete action set, the optimal matching strategy is learned through the deep Q network algorithm, and the matching weight value and matching strategy of each candidate matching pair are output.

8. The instant retail intelligent distribution scheduling method based on spatiotemporal clustering according to claim 7 is characterized in that: Obtaining delivery feedback information, mapping and storing the delivery feedback information with the rider information, and updating the configuration parameters of the matching algorithm according to the delivery feedback information, including: A spatiotemporal alignment algorithm is used to update delivery feedback information to the current rider's historical delivery records, generating a rider profile that includes historical delivery performance, abnormal event records, and capability assessments. and extracting feedback features including route deviation rate, timeliness achievement, service score fluctuation, and exception handling timeliness from the delivery feedback information; The weight coefficients, priority thresholds and path planning parameters in the deep Q network algorithm are optimized based on the feedback characteristics.

9. A computer-readable storage medium storing computer program instructions, characterized in that: The computer program instructions implement the method according to any one of claims 1 to 8 when executed by a processor.

10. An instant retail intelligent distribution scheduling system based on spatiotemporal clustering, characterized by: The method according to any one of claims 1 to 8.

Images