A multimodal collaborative distribution scheduling system and method

Abstract

The application provides a multimodal collaborative distribution scheduling method, comprising: constructing a dynamic environment model according to terrain data and traffic condition data, extracting distribution demand features from real-time order information, adopting a task decomposition technology to split a complex order into a plurality of subtask units to obtain a decomposed subtask set; extracting subtask features of adjacent regions and similar time windows from the optimized subtask sequence, adopting an intelligent combination technology to perform clustering analysis on the subtasks, determining the spatiotemporal correlation between the subtasks, and generating a preliminary subtask aggregation group; calculating path cost and time cost of each distribution batch according to a final resource allocation scheme and the subtask aggregation group, and adopting a genetic algorithm to optimize and iterate the batch path to obtain a time table and route planning of an efficient distribution batch.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of information technology, and particularly relates to a multi-modal collaborative distribution scheduling system and method. BACKGROUND

[0002] Problem background:

[0003] The distribution resource management field occupies a crucial position in the modern logistics system, which is directly related to the improvement of logistics efficiency, cost control and customer experience.

[0004] With the rapid growth of e-commerce and instant delivery demand, how to efficiently integrate various distribution resources has become a key issue for the development of the industry.

[0005] The rise of unmanned distribution technologies such as drones and unmanned vehicles has injected new vitality into the logistics industry, but it has also brought complexity in resource coordination and task optimization.

[0006] The current solutions are mostly focused on the management of a single distribution mode, such as scheduling for manual distribution or unmanned vehicles, lacking unified control and collaborative optimization of multi-modal distribution resources.

[0007] This fragmented management approach is difficult to cope with diversified distribution scenarios, resulting in low resource utilization and limited distribution efficiency.

[0008] The limitations of existing methods are that they often cannot flexibly adapt to different terrain features, traffic conditions and the heterogeneity of distribution resources.

[0009] Traditional scheduling systems rely on static rules or simple algorithms, making it difficult to achieve dynamic decomposition and efficient aggregation of tasks, especially when faced with complex distribution demands, the allocation of sub-tasks is scattered, lacking intelligent integration, resulting in waste of time and resources.

[0010] The core challenges focus on three technical factors: first, the centralized management and coordination mechanism of multi-modal distribution resources is not mature, second, the task decomposition technology for complex scenarios lacks adaptability, and third, the optimization means of sub-task aggregation is insufficient to meet the combined needs of similar areas or similar times.

[0011] Due to the ineffective solution of these technical factors, there is often a contradiction between resource idling and task accumulation in the distribution process, especially in peak periods or complex terrain areas, it is difficult to generate efficient distribution batches, which affects the overall logistics efficiency.

[0012] Therefore, how to develop a multi-modal distribution resource unified operation control platform, realize centralized management of unmanned aerial vehicles, unmanned vehicles and manual distribution, and intelligently combine sub-tasks on the basis of dynamically adapting to terrain and traffic conditions through the two-way optimization technology of task decomposition and aggregation, to form efficient distribution batches, has become a key problem that needs to be solved in this research. SUMMARY

[0013] The application provides a multi-modal collaborative distribution scheduling method, mainly comprising:

[0014] Obtain terrain data, traffic condition data and real-time order information in the distribution area, identify the available resource range of unmanned distribution technology and manual distribution collaboration through multi-modal distribution mode, and determine the initial state and position distribution of each resource in the current period;

[0015] According to the terrain data and traffic condition data, a dynamic environment model is constructed, the distribution demand characteristics are extracted from the real-time order information, the complex order is split into a plurality of sub-task units by using a task decomposition technology, and a sub-task set after decomposition is obtained;

[0016] For the sub-task set after decomposition, the geographical position, time window and resource demand attribute of each sub-task are obtained, the terrain traffic condition is matched through a dynamic adaptability algorithm, if the sub-task conflicts with the current environmental condition, the priority and execution order of the sub-task are adjusted, and an optimized sub-task sequence is obtained;

[0017] The sub-task characteristics of adjacent areas and similar time windows are extracted from the optimized sub-task sequence, the sub-tasks are analyzed by using an intelligent combination technology, the spatio-temporal correlation between the sub-tasks is determined, and a preliminary sub-task aggregation grouping is generated;

[0018] The total resource demand and distribution path constraint of the preliminary sub-task aggregation grouping are obtained, the resource combination of unmanned distribution technology and manual distribution collaboration is allocated through a resource centralized management module, if the resource demand exceeds the current capacity, the supplementary is adjusted from the standby resource pool, and a final resource allocation scheme is obtained;

[0019] According to the final resource allocation scheme and the sub-task aggregation grouping, the path cost and time cost of each distribution batch are calculated, the batch path is optimized and iterated by using a genetic algorithm, and the time table and route planning of the efficient distribution batch are obtained;

[0020] The real-time state of unmanned distribution technology and manual distribution collaboration is updated through the time table and route planning of the efficient distribution batch, the feedback data is collected from the distribution execution process, if the feedback data indicates that the resource is idle or the task is accumulated, the dynamic adaptability algorithm is triggered to re-adjust the task decomposition and aggregation, and a new batch scheduling result is obtained;

[0021] After obtaining the new batch scheduling result, the running state and position information of all distribution resources are synchronously updated by the resource centralized management module, the task decomposition and aggregation input data of the next period are generated, and the continuous optimization capability of the system in the dynamic environment is determined.

[0022] The application also provides a multimodal collaborative distribution scheduling system, characterized by comprising:

[0023] The multimodal data acquisition module is configured to acquire terrain data, traffic condition data and real-time order information in a distribution area, obtain a traffic path distribution through a geographic information system, analyze order-intensive areas and resource demand distribution by K-means algorithm clustering, identify the scope of available resources for unmanned distribution and collaborative resource data for manual distribution, determine the initial state, position distribution and collaborative resource allocation scheme of unmanned distribution equipment and manual distribution, and calculate the optimal path distribution by using Dijkstra algorithm.

[0024] The dynamic environment modeling module is configured to fuse terrain data and traffic conditions to build a dynamic model, adjust the model by real-time updating technology to determine the traffic condition change trend, extract distribution demand features from order information and perform priority sorting, split complex orders into a sub-task set by task decomposition technology, optimize sub-task division and verify task execution sequence by data fusion technology.

[0025] The task decomposition and optimization module is configured to obtain the geographic position, time window and resource demand attribute of a sub-task, match terrain and traffic data by dynamic adaptability algorithm to determine the conflict between the task and the environmental condition, adjust the sub-task priority and execution order if there is a conflict, optimize resource allocation by combining the greedy algorithm, verify the matching consistency of the geographic position and terrain and traffic, and generate the final optimized sub-task sequence.

[0026] The sub-task clustering and aggregation module is configured to extract the adjacent area and time window features of a sub-task, perform clustering analysis by using intelligent combination technology and K-means algorithm, determine the spatio-temporal correlation between sub-tasks, and generate refined sub-task aggregation groups.

[0027] The resource centralized management module is configured to obtain the total resource demand and distribution path constraints of the sub-task aggregation groups, integrate the resource combination of unmanned distribution technology and manual distribution collaboration, supplement from the standby resource pool if the resource demand exceeds the current capacity, optimize the final resource allocation scheme by linear programming algorithm, and monitor the scheme execution state in real time.

[0028] The path and time optimization module is configured to calculate the path cost and time cost according to the resource allocation scheme and the sub-task aggregation groups, optimize the batch path and time table by using genetic algorithm for iterative optimization, and generate the time table and route planning of efficient distribution batches.

[0029] Dynamic feedback adjustment module: configured to collect delivery execution feedback data through sensors and log systems, if resource idling or task accumulation is detected, activate dynamic adaptive algorithm to re-adjust task decomposition and aggregation, generate new batch scheduling results, synchronize update of delivery resource running state and position information, generate task decomposition and aggregation input data for the next cycle, and predict the continuous optimization ability of the system in the dynamic environment through regression algorithm.

[0030] The technical scheme provided by the embodiment of the application can include the following beneficial effects:

[0031] The application discloses an intelligent scheduling method for unmanned delivery and manual delivery cooperation, which can identify available resource range through multi-modal delivery mode, construct a dynamic environment model, and split complex orders into sub-task units by using task decomposition technology. For the sub-task set, the application uses a dynamic adaptive algorithm to match terrain traffic conditions and adjust task priority and execution order. Then, the application uses intelligent combination technology to perform clustering analysis on the sub-tasks and generates preliminary aggregation groups. The application also allocates unmanned and manual delivery resources through a resource centralized management module, optimizes batch paths by using a genetic algorithm, and obtains an efficient delivery schedule and route planning. During execution, the application continuously collects feedback data and triggers a dynamic adjustment mechanism to re-optimize task decomposition and aggregation, thereby achieving continuous optimization ability in a dynamic environment. The method effectively improves delivery efficiency and resource utilization in complex order scenarios, and provides an intelligent solution for the unmanned and manual delivery cooperation mode. BRIEF DESCRIPTION OF DRAWINGS

[0032] Fig. 1 A flowchart of a multi-modal cooperative delivery scheduling method of the application.

[0033] Fig. 2 A schematic diagram of a multi-modal cooperative delivery scheduling method of the application.

[0034] Fig. 3 Another schematic diagram of a multi-modal cooperative delivery scheduling method of the application. DETAILED DESCRIPTION

[0035] The technical scheme of the application will be described below in conjunction with the embodiments. Obviously, the described embodiments are only part of the embodiments of the application, rather than all the embodiments. Based on the embodiments in the application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of the application.

[0036] Embodiment 1

[0037] A multi-modal cooperative delivery scheduling system, comprising:

[0038] Multi-modal data acquisition module: configured to acquire terrain data, traffic condition data and real-time order information within the delivery area, process the traffic path distribution through a geographic information system, cluster analyze the order-intensive area and resource demand distribution combining the K-means algorithm, identify the scope of available resources for unmanned delivery and the collaborative resource data for manual delivery, determine the initial state, location distribution and collaborative resource allocation scheme of unmanned delivery equipment and manual delivery, and calculate the optimal path distribution using Dijkstra algorithm;

[0039] Dynamic environment modeling module: configured to fuse terrain data and traffic conditions to build a dynamic model, adjust the model to determine the traffic condition change trend through real-time updating technology, extract delivery demand features from order information and prioritize, use task decomposition technology to split complex orders into a set of sub-tasks, optimize sub-task division and verify task execution sequence through data fusion technology;

[0040] Task decomposition and optimization module: configured to obtain the geographical location, time window and resource demand attributes of sub-tasks, match terrain and traffic data through dynamic adaptability algorithm to determine the conflict between tasks and environmental conditions, adjust sub-task priority and execution order if there is a conflict, optimize resource allocation combining the greedy algorithm, verify the matching consistency of geographical location and terrain traffic, and generate the final optimized sub-task sequence;

[0041] Sub-task clustering and aggregation module: configured to extract the adjacent area and time window features of sub-tasks, use intelligent combination technology and K-means algorithm for cluster analysis to determine the spatio-temporal correlation between sub-tasks, and generate refined sub-task aggregation groups;

[0042] Resource centralized management module: configured to obtain the total resource demand and delivery path constraints of sub-task aggregation groups, integrate the resource combination of unmanned delivery technology and manual delivery collaboration, supplement from the standby resource pool if the resource demand exceeds the current capacity, optimize the final resource allocation scheme through linear programming algorithm, and monitor the scheme execution status in real time;

[0043] Path and time optimization module: configured to calculate path cost and time cost according to resource allocation scheme and sub-task aggregation groups, use genetic algorithm to optimize and iterate batch path and time table, and generate efficient delivery batch time table and route planning;

[0044] The dynamic feedback adjustment module is configured to collect delivery execution feedback data through sensors and a log system, and if it is detected that resources are idle or tasks are accumulated, a dynamic adaptive algorithm is activated to re-adjust task decomposition and aggregation, generate a new batch scheduling result, synchronously update the running state and position information of the delivery resources, generate input data for task decomposition and aggregation in the next period, and predict the continuous optimization capability of the system in a dynamic environment through a regression algorithm.

[0045] Embodiment 2

[0046] As Figs. 1-3 The multi-modal collaborative delivery scheduling method of the embodiment can specifically include:

[0047] In step S101, terrain data, traffic condition data and real-time order information in the delivery area are obtained, available resource ranges of unmanned delivery technology and manual delivery collaboration are identified through multi-modal delivery mode recognition, and the initial state and position distribution of each resource in the current period are determined.

[0048] [1] Obtain terrain data and traffic condition data in the delivery area, and obtain the traffic path distribution in the area through a preset geographic information system. Obtain the order-intensive area and resource demand distribution by clustering analysis using the K-means algorithm based on the traffic path distribution and real-time order information. If the traffic condition data of the order-intensive area exceeds a preset threshold, identify the available resource range of unmanned delivery through multi-modal analysis. Determine the initial state and position distribution of the unmanned delivery device according to the available resource range of unmanned delivery and the distribution of real-time orders. Obtain collaborative resource data for manual delivery, and determine the initial state and position distribution of manual delivery based on the demand distribution of the order-intensive area. Obtain the collaborative resource allocation scheme of unmanned delivery and manual delivery in the current period through superposition analysis of the position distribution and the initial state. Calculate the optimal path distribution of each resource from the initial state to the target order based on the collaborative resource allocation scheme using the Dijkstra algorithm.

[0049] In a possible implementation, when obtaining terrain data and traffic condition data in the delivery area, satellite images and real-time traffic flow information can be integrated through a geographic information system.

[0050] For example, in a certain city delivery area, high-resolution satellite maps are used to obtain road width, slope and other information, and traffic camera data is used to analyze traffic density.

[0051] Preferably, the geographic information system can superimpose these data to generate a regional traffic path distribution map, clearly showing the traffic capacity of main roads, secondary roads and alleys. This helps to identify which paths are suitable for unmanned delivery devices to travel, significantly improving path planning efficiency.

[0052] Specifically, by combining real-time order information with traffic path distribution, K-means algorithm is used for clustering analysis, which can effectively divide the order-intensive areas.

[0053] For example, there are 100 orders in a certain area at 10 am, and the system clusters them into 3 intensive areas according to order coordinates and path distribution, with 60% of orders in the business district. This analysis can intuitively reflect the distribution of resource demand and provide a basis for distribution resource allocation.

[0054] It should be noted that K-means algorithm ensures the accuracy of area division by iterative optimization and ensures that order points are assigned to the nearest cluster center.

[0055] In one embodiment, if the traffic condition data of the order-intensive area exceeds the preset threshold, such as the speed of the main road in the business district being less than 10 km / h, the multi-modal analysis is started to identify available resources for unmanned distribution.

[0056] For example, the system integrates data of drones, unmanned vehicles and intelligent storage cabinets to confirm that there are 5 drones and 10 storage cabinets available in the business district. This analysis can quickly match resources and demand and avoid traffic congestion affecting distribution efficiency.

[0057] It can be understood that accurate identification of the range of unmanned distribution resources helps to reduce the delay rate of distribution.

[0058] For example, according to the range of unmanned distribution resources and order distribution, the initial state and position distribution of unmanned distribution equipment can be determined, and unmanned drones can be deployed near storage cabinets close to order-intensive areas. Assuming that there are 3 storage cabinets in the business district, 2 drones are deployed at each, and the initial state is set to standby. This distribution can shorten the response time and improve equipment utilization.

[0059] Preferably, the system will also dynamically adjust the equipment state according to the order priority to ensure that high-priority orders are processed first.

[0060] In one possible implementation, when obtaining manual distribution coordination resource data, the real-time position and capacity of the delivery personnel can be analyzed in combination with the demand of the order-intensive area.

[0061] For example, the system detects that an additional 10 delivery personnel are needed in the business district, and there are currently 5 people on standby nearby, and another 5 people can be deployed from adjacent areas.

[0062] It should be noted that this judgment of initial state and position distribution can balance the collaboration efficiency of manual and unmanned distribution and avoid resource waste.

[0063] Specifically, by superimposing the location distribution of unmanned delivery and manual delivery, the collaborative resource allocation scheme can be generated, and the unmanned aerial vehicle can be preferentially allocated to the traffic congestion area, and the manual delivery personnel can be responsible for the complex terrain area.

[0064] For example, the main road of the business district is congested, the unmanned aerial vehicle is responsible for the high-rise office order, and the delivery personnel is responsible for the old community order. This scheme can fully exert the advantages of both and improve the overall delivery efficiency.

[0065] In one embodiment, the Dijkstra algorithm is used to calculate the optimal path distribution for the collaborative resource allocation scheme.

[0066] For example, the unmanned aerial vehicle starts from the storage cabinet to the order point, and the system calculates the shortest flight path based on the path length and obstacle avoidance requirements; the delivery personnel plans the walking or electric vehicle path according to the road trafficability and order point distance.

[0067] It can be understood that this path optimization can significantly reduce the delivery time and energy consumption.

[0068] For example, an order point is 500 meters away from the storage cabinet, and the unmanned aerial vehicle takes 2 minutes to fly directly, while the delivery personnel takes 10 minutes to detour, and the overall efficiency is improved by 30% after optimizing the path. This method ensures efficient use of resources and brings higher stability and customer satisfaction to the delivery business through accurate calculation.

[0069] In step S102, a dynamic environment model is constructed according to terrain data and traffic condition data, delivery demand characteristics are extracted from real-time order information, task decomposition technology is used to split complex orders into several sub-task units, and a decomposed sub-task set is obtained.

[0070] The dynamic model is constructed by fusing terrain data and traffic conditions to obtain environmental feature description. Real-time updating technology is used to adjust the dynamic model to determine the traffic condition change trend. The delivery demand characteristics are extracted from the order information to obtain the demand priority ranking. The demand priority ranking is decomposed into sub-tasks to obtain the initial sub-task division. If there is overlap in the sub-task division, the data fusion technology is used to adjust the sub-task set to determine the optimized sub-task set. The final delivery task allocation is obtained according to the optimized sub-task set division unit. The final delivery task allocation is verified by a machine learning algorithm to determine the task execution sequence.

[0071] Specifically, the dynamic model is constructed by fusing terrain data and traffic conditions to obtain environmental feature description.

[0072] For example, in the delivery area, the terrain data includes slope, road type, etc., and the traffic conditions include traffic flow and intersection congestion degree.

[0073] It can be understood that when fusing these data, the terrain data can be first converted into a two-dimensional grid, each grid being marked with height and road surface properties, and then the real-time heat map of traffic conditions is superimposed to form a dynamic model containing environmental characteristics.

[0074] For example, the main road in a certain area is flat but has heavy traffic, and the model will mark it as a high-passing cost area, while the small road is narrow but smooth, and will be marked as a low-cost path. This model intuitively reflects the impact of the environment on distribution and is convenient for subsequent decision-making. Real-time updating technology is used to adjust the dynamic model to determine the trend of traffic conditions.

[0075] Specifically, traffic information can be updated every 5 minutes through roadside sensors or navigation data.

[0076] In one embodiment, the model dynamically increases the passing cost of a road section due to increased traffic during the morning peak, and predicts the congestion trend in the next 30 minutes.

[0077] Preferably, in combination with historical data analysis, the model can also identify regular changes, such as fixed congestion points from 7:00 to 9:00 on weekdays. This real-time adjustment ensures that distribution decisions are always based on the latest environmental information. The distribution demand characteristics are extracted from the order information to obtain the demand priority ranking.

[0078] For example, order data includes delivery time window, cargo weight, and customer type. One possible implementation is to set the priority of urgent orders (such as fresh food) to the highest, and the priority of ordinary orders to the next.

[0079] For example, a customer orders to be delivered within 1 hour, and the system will automatically increase its priority and mark it as a high-time-efficiency demand. This ranking clearly divides the task urgency and provides a basis for resource allocation. The demand priority ranking is used to decompose the task to obtain a preliminary sub-task division.

[0080] It can be understood that high-priority orders can be decomposed into independent sub-tasks, and low-priority orders can be combined for processing.

[0081] In one embodiment, 3 of the 10 orders are high-priority, and the system assigns them to separate delivery personnel, and the remaining 7 are combined into a regional task. This decomposition reduces resource waste and improves the response speed of critical orders. If there is an overlap in sub-task division, the sub-task set is adjusted through data fusion technology to determine the optimized sub-task set.

[0082] Specifically, overlap refers to multiple sub-tasks covering the same area or time window.

[0083] Exemplarily, two delivery personnel are both assigned to the same cell task, the system will merge the order address and time requirements into one task, and reassign resources. This adjustment avoids repeated delivery and improves overall efficiency. According to the optimized sub-task set division unit, the final delivery task allocation is obtained.

[0084] For example, the system divides the optimized task into the smallest execution unit according to the region and time window.

[0085] In an embodiment, there are 5 sub-tasks in a certain area, and the system assigns them to 3 delivery personnel, each of whom is responsible for the adjacent unit. This division ensures balanced task distribution and reduces the invalid movement of delivery personnel. The final delivery task allocation is verified by a machine learning algorithm to determine the task execution sequence.

[0086] Preferably, a decision tree model can be used to analyze the rationality of task allocation.

[0087] In a possible implementation, the model inputs the task allocation scheme and combines historical delivery data to predict the completion time of each task.

[0088] Exemplarily, a certain allocation scheme shows that a delivery personnel needs to cross 3 regions in succession, and the model suggests that the sequence is inefficient and recommends adjusting it to a near-first sequence. This verification ensures that the task sequence is more in line with the actual execution requirements.

[0089] Step S103, for the decomposed sub-task set, the geographical location, time window and resource demand attributes of each sub-task are obtained, the terrain traffic conditions are matched through a dynamic adaptability algorithm, if the sub-task conflicts with the current environmental conditions, the priority and execution order of the sub-task are adjusted, and the optimized sub-task sequence is obtained.

[0090] The geographical location, time window and resource demand attributes in the sub-task set are obtained, and the terrain traffic and environmental condition data are obtained through database query. The dynamic adaptability algorithm is used to match the geographical location and terrain traffic data to determine whether the sub-task conflicts with the environmental conditions, and the conflict judgment result is obtained. If the conflict judgment result is true, the priority is adjusted according to the time window, and the adjusted priority sequence is obtained. According to the adjusted priority sequence, the execution order is rearranged to obtain the preliminary sorted sub-task sequence. The preliminary sorted sub-task sequence is analyzed through the resource demand attribute, and the greedy algorithm is used to optimize the resource allocation to obtain the resource-optimized sub-task sequence. For the resource-optimized sub-task sequence, the matching consistency of the geographical location and the terrain traffic is verified to obtain the final optimized sequence. Through the final optimized sequence, the execution order of the sub-task set is updated to determine the task scheduling scheme.

[0091] Exemplarily, the geographical location, time window and resource demand attributes in the sub-task set can be obtained through database query.

[0092] In one possible implementation, assuming a city delivery scenario, the subtask set contains multiple delivery points, geographical locations represented as latitude and longitude coordinates, such as delivery point A located at East 120.5, North 30.2, time window from 9:00 AM to 10:00 AM, and resource requirement of 2 delivery personnel and 1 electric vehicle. The database stores the city's terrain data, such as slope, road width, and real-time traffic conditions, such as road congestion index. Through SQL queries, these attributes and environmental data can be quickly extracted, ensuring clear data structure for subsequent processing.

[0093] It should be noted that the dynamic adaptability algorithm matches the geographical location and terrain traffic data to determine whether the subtask conflicts with the environmental conditions, and the core is environmental adaptability.

[0094] In one embodiment, delivery point A is located in a steep slope area, and the terrain data shows that the slope is 15%, while the maximum climbing ability of the delivery vehicle is 10%. The algorithm detects this conflict and marks it as infeasible. In contrast, delivery point B is located in a flat area with no conflict. This matching ensures the executability of the task by comparing the geographical location of the subtask with the terrain restrictions.

[0095] Specifically, if the conflict judgment result is true, the priority is adjusted according to the time window.

[0096] For example, delivery point A cannot be completed on time due to terrain conflict, but its time window is relatively loose and can be adjusted to a lower priority. After adjustment, the priority sequence is advanced to delivery point B in the flat area, and the task with a time window of 9:30-10:30 is executed first. This adjustment optimizes task scheduling by analyzing the flexibility of the time window.

[0097] In one embodiment, after rearranging the execution order, a preliminary sorted subtask sequence is obtained. Assuming the sequence is B, C, A, where C is another flat area delivery point. Based on resource requirement attribute analysis, delivery points B and C each require 1 vehicle, while A requires 2 vehicles. The greedy algorithm optimizes resource allocation, prioritizing vehicle allocation for B and C to maximize resource utilization.

[0098] Preferably, if the total number of vehicles is 2, the algorithm will delay the execution time of A to generate a resource-optimized sequence.

[0099] It can be understood that the matching consistency of geographical location and terrain traffic is verified to ensure the feasibility of the optimized sequence.

[0100] For example, recheck the sequence B, C, A, confirm that the roads of B and C are smooth, and the slope problem of A is solved by replacing a high-power vehicle. After the final optimized sequence is obtained, the execution order of the subtask set is updated to form a task scheduling scheme.

[0101] For example, the delivery point B is completed before 9:30, C is completed at 10:00, and A is arranged to use a special vehicle at 10:30. This scheme is verified by multiple dimensions to ensure efficient scheduling and adapt to environmental changes.

[0102] In step S104, the subtask features of adjacent regions and similar time windows are extracted from the optimized subtask sequence, intelligent combination technology is used for cluster analysis of the subtasks, the spatio-temporal correlation between the subtasks is determined, and a preliminary subtask aggregation group is generated.

[0103] Task feature data is obtained from the subtask sequence, and feature extraction technology is used for processing to obtain a feature set. Adjacent region and time window information is extracted from the feature set to determine the distribution of region-similar and time-similar subtasks. For region-similar and time-similar data, intelligent combination technology is used for grouping to obtain a preliminary clustering result. The preliminary clustering result is analyzed for spatio-temporal correlation to determine the correlation strength between the subtasks, and an optimized clustering group is determined. The optimized clustering group is obtained, and the K-means algorithm is used for secondary clustering of the subtask sequence to obtain a refined aggregation group. The matching degree of task features and spatio-temporal correlation is analyzed through the refined aggregation group, the consistency of the group is judged, and the final subtask aggregation group is obtained. Task optimization information is extracted from the final subtask aggregation group to determine the execution order and resource allocation scheme between the subtasks.

[0104] Illustratively, in the scheduling of logistics distribution tasks, obtaining task feature data of the subtask sequence is the basis for optimizing distribution efficiency. Task feature data usually includes information such as the location of the distribution point, the order quantity, and the distribution time requirement. Feature extraction techniques are used to process these data, such as extracting spatial distribution features from distribution point coordinates or time intensity features from order timestamps through cluster analysis.

[0105] In one possible implementation, assuming that a city has 100 distribution points, each with latitude and longitude and order submission time, the feature extraction technique can divide these points into 5 main clusters by region and identify that the peak period is concentrated from 9am to 11am. The resulting feature set provides a structured input for subsequent analysis.

[0106] It should be noted that when extracting adjacent region and time window information from the feature set, the purpose is to identify subtasks with similar characteristics.

[0107] For example, based on spatial distance calculation, distribution points within 2 kilometers are considered adjacent regions; based on time analysis, orders with overlapping distribution time windows of more than 30 minutes are considered time-similar.

[0108] In one embodiment, 8 out of 10 delivery points in a certain area are located in the same business district, and the order time window is 2-4 pm, which indicates that they are highly similar in space and time. This distribution information provides the basis for grouping.

[0109] Specifically, for regionally similar and time-similar data, intelligent combination technology can group through a rule engine or a machine learning model.

[0110] Preferably, the rule engine can set priorities, such as prioritizing the combination of sub-tasks with close spatial distances, and then considering the degree of overlap of time windows.

[0111] For example, the sub-tasks of the above-mentioned 8 delivery points are combined into one group, while the other two remote points are assigned to other groups. This preliminary clustering result reduces the invalid movement in delivery.

[0112] In one embodiment, spatio-temporal correlation analysis is used to determine the correlation strength between sub-tasks. The correlation strength can be measured by the traffic accessibility of the delivery points or the time dependence of the orders.

[0113] For example, the 8 delivery points in the business district are connected by the main road, and the traffic time is less than 10 minutes, so the correlation strength is high; while the traffic time of the remote point may exceed 30 minutes, so the correlation strength is low. Based on this, the optimized clustering grouping will preferentially retain the combination of sub-tasks with high correlation degree, ensuring that the delivery path is more compact.

[0114] It can be understood that the K-means algorithm is used for secondary clustering of the sub-task sequence to further refine the grouping. The K-means algorithm will reassign the sub-tasks to more optimal clusters based on task characteristics such as distance, time window, order volume, etc.

[0115] For example, the business district sub-tasks in the preliminary grouping may be subdivided into two clusters, one near the shopping center and one near the office building. This refined aggregation grouping improves the accuracy of task allocation.

[0116] For example, by analyzing the matching degree of task characteristics and spatio-temporal correlation through refined aggregation grouping, it can be determined whether the grouping is reasonable.

[0117] In one possible implementation, it is checked whether the average distance of the sub-tasks in each cluster is less than 1 km and whether the time window overlap rate is more than 80%. If the sub-tasks in a certain cluster are too dispersed, the grouping is readjusted. This consistency verification ensures the practicality of the grouping.

[0118] Preferably, task optimization information is extracted from the final sub-task aggregated grouping, such as the center point of each cluster as the delivery starting point, or determining the priority delivery order based on the time window.

[0119] In one embodiment, the sub-tasks of the business district are arranged on the same delivery vehicle, with the starting point set at the shopping center and the sequence arranged according to the time window from early to late. This optimization information is directly translated into a specific execution sequence and resource allocation scheme, for example, assigning larger-capacity vehicles to high-density areas. This way significantly improves delivery efficiency.

[0120] In step S105, the total resource demand and delivery path constraints of the preliminary sub-task aggregation group are obtained, and the resource combination of unmanned delivery technology and manual delivery collaboration is allocated by the resource centralized management module. If the resource demand exceeds the current capacity, supplementary resources are allocated from the standby resource pool to obtain the final resource allocation scheme.

[0121] The resource demand and delivery path constraints of the sub-task aggregation group are obtained by the task decomposition module, the preliminary demand distribution data is determined, the delivery path constraints are processed using a path optimization algorithm to obtain an optimized path allocation scheme, the resource combination of unmanned delivery technology and manual delivery collaboration is integrated by the resource centralized management module to determine the feasibility of the allocation. If the demand exceeds the current capacity limit, the excess part is analyzed by the capacity evaluation tool to obtain available resource data from the standby resource pool. The supplementary resources are allocated from the standby resource pool to the resource combination allocation to obtain the adjusted resource allocation result. The adjusted resource allocation result is optimized using a linear programming algorithm to determine the final allocation scheme. The execution status of the final allocation scheme is verified by the real-time monitoring module to obtain dynamic adjustment data during execution.

[0122] Illustratively, when the resource demand and delivery path constraints of the sub-task aggregation group are obtained by the task decomposition module, the specific task demand in the unmanned delivery scenario can be analyzed first. Suppose there are 10 delivery points in a certain city area, each requiring different number of packages, such as 50 pieces for point A and 30 pieces for point B. The task decomposition module will organize these points into sub-task groups according to geographical location and package quantity, and record the delivery path constraints of each group, such as some areas that need to be detoured due to traffic restrictions. This decomposition method can clearly present the demand distribution, facilitating subsequent optimization.

[0123] In one possible implementation, when the path optimization algorithm is used to process the delivery path constraints, an optimized path allocation scheme can be generated based on actual geographic data.

[0124] For example, for the above-mentioned 10 delivery points, the algorithm will preferentially select paths with shorter distance and avoid restricted areas. Suppose the straight-line distance from point A to point B is 5 kilometers, but due to restrictions, it needs to be detoured for 7 kilometers. The algorithm will calculate the time and energy cost and recommend an alternative path of 6 kilometers. This method ensures that the path is efficient and meets the actual constraints.

[0125] Specifically, when the resource centralized management module integrates the resource combination of unmanned delivery technology and manual delivery coordination, the module can allocate unmanned vehicles and manual delivery personnel according to the task volume.

[0126] For example, among the 50 packages at point A, 40 are delivered by unmanned vehicles, and 10 are manually processed due to special requirements. The module will evaluate the load capacity of the unmanned vehicles and the delivery efficiency of the manual personnel to determine whether the combination is feasible. If the unmanned vehicles are fully loaded, the module will dynamically adjust the proportion of manual delivery. This integration method can flexibly respond to different task requirements.

[0127] It should be noted that if the demand exceeds the current capacity limit, the capacity evaluation tool will analyze the excess and obtain data from the backup resource pool.

[0128] For example, the 50 packages at point A exceed the maximum capacity of 40 packages for the unmanned vehicles, so the tool will query the backup resource pool to find an additional unmanned vehicle or temporary manual delivery personnel. The analysis process will consider the scheduling time and cost of the backup resources to ensure that the additional resources are in place in a timely manner.

[0129] Preferably, after the additional resources from the backup resource pool are allocated to the resource combination, an adjusted resource allocation result can be obtained.

[0130] For example, after an additional unmanned vehicle is allocated, the delivery requirements at point A are met, and the allocation result is updated to two unmanned vehicles delivering 40 and 10 packages, and a manual delivery personnel responsible for the remaining part. This allocation method can quickly respond to unexpected demand.

[0131] In one embodiment, when the linear programming algorithm is used to optimize the adjusted resource allocation result, the final scheme can be determined by simulating different delivery combinations.

[0132] For example, the algorithm compares the efficiency of unmanned vehicle full-automatic delivery and unmanned vehicle plus manual mixed delivery, and preferentially selects the scheme with shorter time and lower cost. Assuming that mixed delivery can save 20% of the delivery time, the algorithm will lock this scheme. This optimization method can improve the overall delivery efficiency.

[0133] It can be understood that when the real-time monitoring module verifies the execution status of the final allocation scheme, dynamic adjustment data can be obtained.

[0134] For example, during the delivery process, if point B is delayed due to temporary road closures, the monitoring module will provide real-time feedback and suggest switching to a backup path or increasing the number of manual delivery personnel. The adjustment data will record the delay time and the effect of the solution, providing a reference for subsequent task optimization. This real-time monitoring method can ensure that the delivery process is stable and controllable.

[0135] Step S106, according to the final resource allocation scheme and the sub-task aggregation grouping, the path cost and time cost of each distribution batch are calculated, the genetic algorithm is used to optimize the iteration of the batch path, and the time table and route planning of the efficient distribution batch are obtained.

[0136] The sub-task groups are divided by the resource allocation scheme, the grouping result after task aggregation is calculated, the basic data of the distribution batch is obtained, the path cost and time cost are extracted from the basic data of the distribution batch, the initial cost value of each batch is determined by the cost calculation method, the path cost is judged whether it exceeds the preset threshold value, if it exceeds, the batch path is adjusted by the genetic algorithm, and the optimized path scheme is obtained, according to the optimized path scheme, the change trend of the time cost is obtained, the time table is iterated by the genetic algorithm, the time sequence of efficient distribution is determined, the route planning of each distribution batch is calculated by the time sequence and the path scheme, and the final route distribution data is obtained, the time cost is judged whether it meets the preset threshold value, if it does not meet, the grouping mode of task aggregation is adjusted, and the optimization iteration process is repeated, according to the adjusted grouping mode, the path cost and time cost are determined by the cost calculation method, and the time table and route planning of the efficient distribution batch are obtained

[0137] Specifically, the sub-task groups are divided by the resource allocation scheme, the grouping result after task aggregation is calculated, and the basic data of the distribution batch is obtained, and the core is to decompose the complex distribution demand into an operable unit.

[0138] For example, a certain city distribution center needs to handle 1000 orders per day, including fresh food, daily necessities, etc.

[0139] Preferably, the tasks can be grouped into 10 sub-task groups according to the geographical position and order type, about 100 orders per group, to ensure balanced grouping. In implementation, batch data can be automatically generated by the order management system based on regional division and category priority, such as a fresh food priority batch in the northern region containing 50 orders, basic data including order quantity, distribution distance, etc. The path cost and time cost are extracted from the basic data of the distribution batch, the initial cost value of each batch is determined by the cost calculation method, and the focus is on quantifying the distribution cost.

[0140] Specifically, the path cost can be calculated based on distance and traffic conditions, and the time cost considers the distribution time.

[0141] For example, the total distance of the northern batch is 200 kilometers, the average cost per kilometer is 0.5 yuan, and the path cost is 100 yuan. The time cost is calculated according to 15 minutes per order and 60 yuan per hour of labor cost, and the time cost is about 750 yuan. In implementation, the cost calculation tool can integrate map data and real-time traffic information to ensure the accuracy of the value. For the initial cost value, it is judged whether the path cost exceeds the preset threshold. If it exceeds, the batch path is adjusted by genetic algorithm to obtain the optimized path scheme, and this process aims to reduce the excessive cost.

[0142] In an embodiment, assuming the threshold value is 80 yuan, the northern batch cost 100 yuan exceeds the threshold. The genetic algorithm can simulate multiple path combinations and iteratively select the shortest path, such as merging some orders to adjacent routes, and the adjusted path cost is reduced to 75 yuan.

[0143] It should be noted that the algorithm needs to balance the path length and vehicle loading, and in implementation, dynamic adjustment can be realized through scheduling software. According to the optimized path scheme, the change trend of the time cost is obtained, and the genetic algorithm is used to iterate the time table to determine the time sequence of efficient distribution, the purpose is to optimize the distribution rhythm.

[0144] For example, the adjusted path may shorten the delivery time of some orders, such as from 15 minutes to 12 minutes. The genetic algorithm further optimizes the time table, prioritizes peak period orders, and ensures overall time efficiency.

[0145] For example, the total time cost of the adjusted batch is reduced to 600 yuan. In implementation, the time sequence can be automatically generated by the scheduling system, combined with the driver scheduling to improve efficiency. Through the time sequence and the path scheme, the route planning of each distribution batch is calculated to obtain the final route distribution data, the core is to form an executable distribution plan.

[0146] In a possible implementation, the northern batch is planned as 3 routes, each covering 30-40 orders, and the route distribution data includes start point, end point and passing point. In implementation, the navigation system can directly import the data, and the driver can execute according to the plan to ensure that all orders are covered. After obtaining the route distribution data, it is judged whether the time cost meets the preset threshold. If it does not meet, the grouping method of task aggregation is adjusted, and the optimization iteration process is repeated, emphasizing dynamic error correction.

[0147] It can be understood that if the time cost still exceeds the threshold of 600 yuan, the batch can be re-grouped, such as splitting the northern batch into two sub-groups, and optimizing the path and time respectively. In implementation, the scheduling center can judge in real time through the monitoring system to quickly iterate the grouping. According to the adjusted grouping method, the path cost and time cost are determined again by the cost calculation method to obtain the time table and route planning of efficient distribution batch, ensuring the optimality of the scheme.

[0148] For example, after re-grouping, the cost of the northern batch is reduced to 70 yuan, and the time cost is 550 yuan. The time table is accurate to the departure time of each order. In implementation, the system can generate a visual report to facilitate the verification of the dispatcher and ensure efficient distribution.

[0149] In step S107, the real-time state of the unmanned distribution technology and the manual distribution cooperation is updated by the time table and the route planning of the efficient distribution batch. Feedback data is collected from the distribution execution process. If the feedback data indicates that the resources are idle or the tasks are accumulated, a dynamic adaptability algorithm is triggered to re-adjust the task decomposition and aggregation, and a new batch scheduling result is obtained.

[0150] The feedback data in the distribution execution process is collected in real time by the sensor and the log system to obtain the current state of the task execution. If the feedback data exceeds the preset resource idle threshold or task accumulation threshold, a dynamic adaptability algorithm is activated. The feedback data is processed by the dynamic adaptability algorithm to obtain a preliminary adjustment scheme of task decomposition and task aggregation. A new batch scheduling arrangement is obtained by the optimized task decomposition result. According to the batch scheduling arrangement, corresponding route planning data is generated. After obtaining the route planning data, the real-time state of the unmanned distribution and the manual distribution is updated to obtain a cooperative execution scheme. The distribution execution process is adjusted by the cooperative execution scheme to obtain new feedback data for cyclic input

[0151] Specifically, the real-time collection of feedback data in the distribution execution process by the sensor and the log system is the basis for building a dynamic distribution system.

[0152] For example, the sensor can include a positioning device of the unmanned distribution vehicle, a load sensor, and an environment perception device, and the log system records the task completion time and abnormal events. Assuming that a distribution center collects data every minute to obtain the current position of the distribution vehicle, the remaining power, and the task completion rate, etc. These data reflect the real-time state of task execution, for example, a vehicle stays in a certain area for too long, indicating that it may encounter traffic congestion or equipment failure.

[0153] In one possible implementation, if the feedback data exceeds the preset resource idle threshold or task accumulation threshold, adjustment needs to be triggered.

[0154] For example, the resource idle threshold is set to be more than 70% of the vehicle empty rate, and the task accumulation threshold is set to be more than 20 orders in a certain area. Assuming that the empty rate of a certain distribution vehicle is 80%, or 25 orders are accumulated in a certain area, the system will automatically mark it as an abnormality and activate the dynamic adaptability algorithm. The core of this algorithm is to quickly respond to abnormal states.

[0155] Preferably, whether the task needs to be re-assigned or the route needs to be adjusted is determined by analyzing the historical data and the current feedback.

[0156] Specifically, when processing feedback data, the dynamic adaptive algorithm generates task decomposition and aggregation schemes based on task priorities and the real-time status of delivery vehicles.

[0157] For example, if orders in a certain area surge, the algorithm can split large orders into multiple small tasks and assign them to nearby idle vehicles, or aggregate small orders from adjacent areas into a batch and assign them to a single vehicle. Suppose a vehicle was originally scheduled to deliver 10 orders, the algorithm adjusts it to split the 10 orders into 5 orders and 5 orders, which are delivered by two vehicles respectively, thereby improving efficiency.

[0158] It should be noted that the system generates new batch scheduling arrangements based on the optimized task decomposition results.

[0159] For example, after adjustment, the task volume of each vehicle is balanced, and the delivery time window is compressed from the original 2 hours to 1.5 hours. The scheduling arrangement also takes into account the vehicle's battery level and road conditions, and prioritizes vehicles with sufficient battery levels for long-distance tasks. Such arrangements maximize resource utilization.

[0160] In one embodiment, when generating route planning data based on batch scheduling arrangements, the system considers real-time traffic conditions and order locations.

[0161] For example, a vehicle needs to deliver 3 orders, located at points A, B, and C respectively. The system selects a path from A to B to C based on traffic conditions to avoid congested areas. Such planning data directly affects delivery efficiency.

[0162] Preferably, buffer time is also reserved to deal with unexpected situations.

[0163] It can be understood that after obtaining the route planning data, the real-time status of unmanned delivery and manual delivery is updated to form a collaborative execution scheme.

[0164] For example, unmanned vehicles are responsible for main road delivery, and manual delivery personnel take over the last mile. Suppose a certain unmanned vehicle completes 80% of its tasks but runs out of power, the system transfers the remaining tasks to manual delivery personnel to ensure timely delivery of orders. Such collaborative schemes improve the flexibility of delivery.

[0165] For example, after adjusting the delivery process through the collaborative execution scheme, the system generates new feedback data and enters the cycle.

[0166] For example, after adjustment, the task backlog in a certain area decreases from 25 orders to 10 orders, and the vehicle empty rate decreases from 80% to 50%. These data re-enter the system to form a closed-loop optimization. This continuous feedback loop ensures that the delivery system always operates efficiently while adapting to complex and changing environments.

[0167] Step S108, after obtaining the new batch scheduling result, the running state and location information of all distribution resources are updated synchronously through the resource centralized management module, the task decomposition and aggregation input data of the next period are generated, and the continuous optimization capability of the system in the dynamic environment is determined.

[0168] After obtaining the batch scheduling result, the running state and location information are updated synchronously through the management module to obtain the latest distribution data of the distribution resources. The task decomposition elements are extracted from the latest distribution data, the clustering algorithm is used to generate the aggregation data, and the task allocation basis of the next period is determined. The change trend of the dynamic environment is analyzed through the aggregation data, and it is judged whether the current optimization capability of the system meets the requirements. If the change trend exceeds the preset threshold, the running state is adjusted through the resource management to obtain the optimized resource distribution data. According to the optimized resource distribution data, the batch scheduling input of the next period is generated, and the priority sequence of task decomposition is determined. The influence of the dynamic environment on the optimization capability is predicted by using the regression algorithm, and the adaptability of the system in the next period is judged. The priority sequence and the aggregation data are combined through the management module to generate the final task allocation scheme.

[0169] Exemplarily, after obtaining the batch scheduling result, the process of the management module synchronously updating the running state and location information can be realized by monitoring the positions of the distribution vehicles and personnel in real time. Assuming that a city distribution network includes unmanned distribution vehicles and manual distribution personnel, the management module obtains the current position, speed and task completion of each vehicle in real time through GPS and Internet of Things devices.

[0170] For example, after a certain unmanned distribution vehicle completes the task in region A, the position is updated to A point standby, and after a manual distribution personnel completes the distribution in region B, the state is updated to accept new tasks. The latest distribution data reflects that there are 2 unmanned vehicles on standby in region A and 3 distribution personnel are idle in region B. This real-time updating method ensures the accuracy of the resource distribution data and provides a reliable basis for subsequent task allocation.

[0171] In a possible implementation, when the task decomposition elements are extracted from the latest distribution data, the geographical location, urgency and resource matching degree of the task can be focused on. Assuming that there are currently 10 to-be-distributed orders distributed in regions C and D, the management module will extract the delivery time requirement and goods type of each order.

[0172] For example, the orders in region C are mostly fresh food and require delivery within 30 minutes, and the orders in region D are ordinary packages with ample time. Based on these elements, the clustering algorithm clusters the orders in region C into one class, and preferentially allocates unmanned vehicles to ensure speed. The orders in region D are allocated to manual distribution personnel to take advantage of their flexibility. This clustering method can effectively match the characteristics of tasks and resources.

[0173] Specifically, after generating aggregated data using clustering algorithms, the process of determining the basis for task allocation in the next cycle can be achieved by analyzing the geographical density of orders and the distribution of resources. Suppose that orders in region C are concentrated in a certain business district, the algorithm will aggregate these orders into a delivery batch and assign a robot car to execute it. While orders in region D are more scattered, the algorithm will split them into multiple small batches and assign them to different delivery personnel. The generation of aggregated data clearly outlines the priority of task allocation, laying the foundation for scheduling in the next cycle.

[0174] It should be noted that when analyzing the trend of dynamic environmental changes, historical data and real-time feedback can be used to judge the optimization ability of the system.

[0175] For example, during a certain delivery peak, the number of orders in region C suddenly increases by 20%, exceeding the carrying capacity of the robot car. Through trend analysis, the management module determines that the system needs to temporarily allocate resources. If the change trend exceeds the preset threshold, for example, if the order backlog exceeds 10 orders, the system will allocate 1 artificial delivery personnel from region B to support region C through the resource management module. The optimized resource distribution data shows that the resource shortage in region C has been alleviated, ensuring the delivery efficiency.

[0176] In one embodiment, when generating input for batch scheduling in the next cycle, the determination of the priority sequence is particularly important. Suppose that the system prioritizes fresh food orders in region C first and ordinary orders in region D second based on the urgency of the orders. The management module generates input data indicating that the robot car should prioritize executing tasks in region C, and the artificial delivery personnel should handle tasks in region D afterwards. This priority sequence ensures that critical tasks are completed first.

[0177] Preferably, when predicting the impact of dynamic environment, a regression algorithm can be used to analyze historical delivery data.

[0178] For example, the system analyzes the delivery records of the past week and finds that rain will cause a 15% increase in delivery time. Based on this, the regression algorithm predicts that if it rains in the next cycle, the system needs to increase resource investment by 10%. This prediction method enhances the adaptability of the system to the external environment.

[0179] For example, when the management module combines the priority sequence with the aggregated data to generate the final task allocation scheme, it can display the allocation results through a visual interface. Suppose that the robot car in region C is assigned 5 fresh food orders, and the 3 delivery personnel in region D each receive 2 ordinary orders. The allocation scheme clearly outlines the routes and time requirements for each delivery personnel, ensuring efficient task execution. This combination makes the allocation scheme both scientific and operational.

[0180] It can be understood that the implementation of each of the above steps constitutes an efficient distribution resource management closed loop. From real-time updating to final distribution, each step is linked together, and the data flows smoothly. The advantage of this method is that it can dynamically respond to changes in distribution demand, maximize resource utilization, and at the same time, ensure the timeliness of the task and customer satisfaction.

[0181] In addition, it should be noted that the various specific technical features described in the above specific embodiments can be combined in any appropriate manner without contradiction. In order to avoid unnecessary repetition, various possible combinations are not described again by the present application. Furthermore, various different embodiments of the present application can be combined in any manner, as long as they do not contradict the idea of the present application, and they should also be considered as disclosed by the present application.

Claims

1. A multi-modal coordinated distribution dispatching method, characterized by, The method comprises: S101, acquiring terrain data, traffic condition data and real-time order information in a distribution area, identifying available resource ranges of unmanned distribution technology and manual distribution collaboration through multi-modal distribution mode, and determining initial states and position distributions of each resource in a current time period; S102, constructing a dynamic environment model according to the terrain data and the traffic condition data, extracting distribution demand features from the real-time order information, adopting task decomposition technology to split a complex order into a plurality of sub-task units to obtain a decomposed sub-task set; S103, for the decomposed sub-task set, acquiring geographical positions, time windows and resource demand attributes of each sub-task, matching terrain traffic conditions through a dynamic adaptability algorithm, judging whether the sub-task conflicts with a current environment condition, adjusting a priority and an execution order of the sub-task if the sub-task conflicts with the current environment condition, and obtaining an optimized sub-task sequence; S104, extracting sub-task features of adjacent areas and similar time windows from the optimized sub-task sequence, adopting intelligent combination technology to perform clustering analysis on the sub-task, determining spatio-temporal correlation between the sub-tasks, and generating a preliminary sub-task aggregation group; S105, acquiring a total resource demand and a distribution path constraint of the preliminary sub-task aggregation group, allocating resource combinations of the unmanned distribution technology and the manual distribution collaboration through a resource centralized management module, judging whether the resource demand exceeds a current capacity, and supplementing from a standby resource pool if the resource demand exceeds the current capacity, and obtaining a final resource allocation scheme; S106, calculating path cost and time cost of each distribution batch according to the final resource allocation scheme and the sub-task aggregation group, adopting a genetic algorithm to optimize and iterate batch paths, and obtaining a time table and a route plan of an efficient distribution batch; S107, updating real-time states of the unmanned distribution technology and the manual distribution collaboration through the time table and the route plan of the efficient distribution batch, collecting feedback data from a distribution execution process, and judging whether the feedback data indicates that resources are idle or tasks are accumulated, and triggering the dynamic adaptability algorithm to re-adjust task decomposition and aggregation to obtain a new batch scheduling result; S108, after obtaining the new batch scheduling result, synchronously updating running states and position information of all distribution resources through the resource centralized management module, generating input data of task decomposition and aggregation in a next period, and determining a continuous optimization capability of the system in a dynamic environment; The S102 comprises: constructing a dynamic model through terrain data and traffic condition fusion to obtain environment feature description; adopting real-time updating technology to adjust the dynamic model to determine a traffic condition change trend; extracting distribution demand features from order information to obtain demand priority ranking; performing task decomposition on the demand priority ranking to obtain preliminary sub-task division; if there is overlap in the sub-task division, adjusting the sub-task set through data fusion technology, and judging an optimized sub-task set; dividing units according to the optimized sub-task set to obtain final distribution task allocation; adopting a machine learning algorithm to verify the final distribution task allocation to determine a task execution sequence; The S103 comprises: acquiring geographical positions, time windows and resource demand attributes in the sub-task set, and obtaining terrain traffic and environment condition data through database query; The geographical position and the terrain traffic data are matched by a dynamic adaptive algorithm, whether the subtask conflicts with the environmental condition is judged, and a conflict judgment result is obtained; If the conflict judgment result is true, the priority is adjusted according to the time window, and an adjusted priority sequence is obtained; According to the adjusted priority sequence, the execution order is rearranged, and a preliminarily ordered subtask sequence is obtained; The preliminarily ordered subtask sequence is analyzed through the resource demand attribute, a greedy algorithm is used to optimize the resource allocation, and a resource-optimized subtask sequence is obtained; For the resource-optimized subtask sequence, the matching consistency of the geographical position and the terrain traffic is verified, and a final optimized sequence is obtained; Through the final optimized sequence, the execution order of the subtask set is updated, and the task scheduling scheme is determined; The S104 comprises: Task feature data is obtained through the subtask sequence, and a feature set is obtained by processing the task feature data by using a feature extraction technology; Adjacent area and time window information is extracted from the feature set, and region-similar and time-similar subtask distribution is determined; For region-similar and time-similar data, grouping is performed by using an intelligent combination technology, and a preliminary clustering result is obtained; The preliminary clustering result is analyzed in time and space, the correlation strength between the subtasks is judged, and an optimized clustering group is determined; The optimized clustering group is obtained, and the subtask sequence is clustered again by using a K-means algorithm, and a refined aggregation group is obtained; The matching degree of the task feature and the time-space correlation is analyzed through the refined aggregation group, the consistency of the group is judged, and a final subtask aggregation group is obtained; Task optimization information is extracted from the final subtask aggregation group, and the execution order and the resource allocation scheme between the subtasks are determined.

2. The method of claim 1, wherein, The S101 comprises: Obtain the terrain data and the traffic condition data in the distribution area, obtain the passing path distribution in the area by processing through a preset geographic information system, obtain the order-intensive area and the resource demand distribution by clustering analysis through the K-means algorithm by combining the passing path distribution with the real-time order information, if the traffic condition data of the order-intensive area exceeds a preset threshold value, identify the available resource range of unmanned distribution through multi-modal analysis, determine the initial state and the position distribution of the unmanned distribution equipment according to the available resource range of unmanned distribution and the distribution of real-time orders, obtain the collaborative resource data of manual distribution, judge the initial state and the position distribution of manual distribution by combining the demand distribution of the order-intensive area, obtain the collaborative resource allocation scheme of unmanned distribution and manual distribution in the current period through superposition analysis of the position distribution and the initial state, and calculate the optimal path distribution of each resource from the initial state to the target order by using the Dijkstra algorithm for the collaborative resource allocation scheme.

3. The method of claim 1, wherein, The S105 comprises: The task decomposition module obtains the resource requirement quantity and the distribution path constraint of the subtask aggregation group, determines the preliminary demand distribution data, processes the distribution path constraint by using a path optimization algorithm, obtains an optimized path allocation scheme, integrates the unmanned distribution technology and the resource combination of manual distribution collaboration by using the resource centralized management module, judges the feasibility of the allocation, and if the demand exceeds the current capacity limit, analyzes the exceeding part by using a capacity evaluation tool, obtains the available resource data of the backup resource pool, allocates the supplementary resource from the backup resource pool to the resource combination allocation, obtains the adjusted resource allocation result, optimizes the adjusted resource allocation result by using a linear programming algorithm, determines the final allocation scheme, verifies the execution state of the final allocation scheme by using the real-time monitoring module, and obtains the dynamic adjustment data in the execution process.

4. The method of claim 1, wherein, The S106 includes: The resource allocation scheme is used to divide the subtask group, the grouping result after task aggregation is calculated, the basic data of the distribution batch is obtained, the path cost and the time cost are extracted from the basic data of the distribution batch, the initial cost value of each batch is determined by using a cost calculation method, the path cost is judged according to the initial cost value, if the path cost exceeds the preset threshold, the batch path is adjusted by using a genetic algorithm, the optimized path scheme is obtained, the change trend of the time cost is obtained according to the optimized path scheme, the time table is iterated by using a genetic algorithm, the time sequence of efficient distribution is determined, the route planning of each distribution batch is calculated by using the time sequence and the path scheme, and the final route distribution data is obtained. The route distribution data is obtained, and it is judged whether the time cost meets the preset threshold. If not, the grouping mode of task aggregation is adjusted, and the optimization iteration process is repeated. The path cost and the time cost are determined again by using the cost calculation method according to the adjusted grouping mode, and the time table and the route planning of efficient distribution batch are obtained.

5. The method of claim 1, wherein, The S107 includes: The sensor and the log system are used to collect feedback data in the distribution execution in real time, the current state of task execution is obtained, if the feedback data exceeds the preset resource idle threshold or the task accumulation threshold, the dynamic adaptability algorithm is activated, the adjustment demand is determined, the feedback data is processed by using the dynamic adaptability algorithm, the preliminary adjustment scheme of task decomposition and task aggregation is obtained, the new batch scheduling arrangement is obtained by using the optimized task decomposition result, the corresponding route planning data is generated according to the batch scheduling arrangement, the real-time state of unmanned distribution and manual distribution is updated after the route planning data is obtained, the collaborative execution scheme is obtained, the distribution execution process is adjusted by using the collaborative execution scheme, and the new feedback data is inputted in a cycle.

6. The method of claim 1, wherein, The S108 includes: After the batch scheduling result is obtained, the management module is used to update the running state and the position information synchronously, and the latest distribution data of the distribution resource is obtained; The task decomposition elements are extracted from the latest distribution data, the clustering algorithm is used to generate aggregation data, and the task allocation basis of the next period is determined; The change trend of the dynamic environment is analyzed by using the aggregation data, and it is judged whether the optimization ability of the system meets the requirement; If the change trend exceeds the preset threshold, the running state is adjusted by using the resource management, and the optimized resource distribution data is obtained; According to the optimized resource distribution data, a batch scheduling input of the next period is generated to determine a priority sequence of task decomposition; A regression algorithm is used to predict the influence of dynamic environment on optimization capability, and the adaptability of the system in the next period is judged; Through the management module, the priority sequence is combined with the aggregated data to generate the final task allocation scheme.

7. A multi-modal coordinated distribution dispatching system, characterized by, The multi-modal collaborative distribution scheduling method of claim 1 comprises: A multi-modal data acquisition module is configured to acquire terrain data, traffic condition data and real-time order information in the distribution area, process the passing path distribution through a geographic information system, cluster analyze the order dense area and resource demand distribution by combining a K-means algorithm, identify the scope of available resources for unmanned distribution and collaborative resource data for manual distribution, determine the initial state, location distribution and collaborative resource allocation scheme of unmanned distribution equipment and manual distribution, and calculate the optimal path distribution by using a Dijkstra algorithm; A dynamic environment modeling module is configured to fuse terrain data and traffic conditions to build a dynamic model, adjust the model through real-time updating technology to determine the traffic condition change trend, extract distribution demand features from order information and perform priority sorting, use task decomposition technology to split complex orders into a sub-task set, optimize sub-task division and verify task execution sequence through data fusion technology; A task decomposition and optimization module is configured to acquire the geographical location, time window and resource demand attributes of sub-tasks, match terrain and traffic data through dynamic adaptability algorithm to judge the conflict between tasks and environmental conditions, adjust sub-task priority and execution order if there is a conflict, optimize resource allocation by combining a greedy algorithm, verify the matching consistency of geographical location and terrain and traffic, and generate the final optimized sub-task sequence; A sub-task clustering and aggregation module is configured to extract the adjacent area and time window features of sub-tasks, use intelligent combination technology and K-means algorithm for cluster analysis to determine the spatio-temporal correlation between sub-tasks, and generate refined sub-task aggregation groups; A resource centralized management module is configured to acquire the total resource demand and distribution path constraints of sub-task aggregation groups, integrate the resource combination of unmanned distribution technology and manual distribution collaboration, supplement from the standby resource pool if the resource demand exceeds the current capacity, optimize the final resource allocation scheme through linear programming algorithm, and monitor the scheme execution state in real time; A path and time optimization module is configured to calculate path cost and time cost according to the resource allocation scheme and sub-task aggregation groups, use genetic algorithm to optimize and iterate batch path and time table, and generate an efficient distribution batch time table and route planning; A dynamic feedback adjustment module is configured to collect distribution execution feedback data through sensors and log systems, activate the dynamic adaptability algorithm to re-adjust task decomposition and aggregation if resource idling or task accumulation is detected, generate new batch scheduling results, update the running state and location information of distribution resources in synchronization, generate input data for task decomposition and aggregation in the next period, and predict the sustained optimization capability of the system in the dynamic environment through a regression algorithm.

Images