A logistics path collaborative optimization method and system

Abstract

The application discloses a kind of logistics path coordination optimization method and system, it is related to logistics technical field.The logistics path coordination optimization method and system of the kind, comprising: S1, receive user order task, real-time collection distribution data and synchronization distribution resource state, the real-time evaluation of task and resource allocation;S2, establish traffic graph network, mark distribution conflict point, generate distribution path, carry out unmanned vehicle and distribution staff path coordination optimization;S3, continuously perceive distribution resource state, carry out the forecast analysis of the availability and execution capability of distribution resource, and implement optimization measures;S4, identify emergent event, carry out task transfer and redistribution, dynamically re-distribute resource and path, carry out redistribution stress evaluation;S5, analysis system log record, construct training data set, evaluate scheduling effect and optimization strategy, and generate report.It solves the problem of distribution resource conflict caused by simultaneously running unmanned distribution vehicle and manual distribution staff in the city.

Description

Technical Field

[0001] This invention relates to the field of logistics technology, specifically to a method and system for collaborative optimization of logistics routes. Background Technology

[0002] With the continuous growth of urban logistics demand, last-mile delivery tasks are characterized by high density, diversity, and high timeliness. Especially in complex scenarios such as commercial districts, residential areas, and office parks, delivery tasks exhibit significant spatial clustering and temporal conflicts. Simultaneously, delivery resources are gradually expanding from traditional human riders to diverse carriers such as intelligent unmanned vehicles and unmanned delivery robots, and the heterogeneity of resource status further complicates scheduling issues. To improve overall logistics efficiency and reduce delivery costs, building an intelligent system capable of multi-task, multi-resource collaborative matching and route optimization in a dynamic environment has become an important development direction for urban smart logistics.

[0003] For example, invention patent CN115187169A discloses a logistics distribution system and method based on collaborative path planning, relating to the field of supply chain logistics technology. The system includes a data acquisition module, a data preprocessing module, a distribution path planning module, and a distribution instruction generation module. First, it acquires warehouse information and customer order information, calculates the distance between nodes and generates an omnidirectional network graph, then assigns customers to warehouses based on order information, solves the collaborative path planning model to generate distribution paths and transfer nodes, calculates item delivery and transfer information, and finally sends distribution instructions to the delivery equipment to deliver items from the warehouse to the customer. This solution can improve the utilization rate of delivery equipment and reduce delivery mileage while meeting delivery requirements, thereby reducing operating costs.

[0004] For example, invention patent CN114358675B discloses a method for multi-drone-multi-truck collaborative logistics delivery route planning, belonging to the field of drone logistics technology. It includes the following steps: Step 1: Establishing a mixed-integer linear programming model for the multi-drone-multi-truck collaborative logistics delivery route planning problem; Step 2: Initially planning truck delivery routes based on the K-Means algorithm and genetic algorithm; Step 3: Designing a route planning search operator, introducing a variable neighborhood search framework to jointly optimize the delivery routes of drones and trucks based on the truck delivery routes, and solving the constructed mixed-integer linear programming model. The method provided by this invention considers the different purchase costs of trucks and drones, reasonably optimizes the purchase and usage of transportation vehicles, effectively reduces the total delivery cost, further optimizes the drone-truck delivery scheme, and makes up for the shortcomings of existing joint delivery models and methods.

[0005] However, in the process of implementing the inventive technical solution in the embodiments of this application, it was found that the above-mentioned technology has at least the following technical problems:

[0006] The matching method is static, making it difficult to adjust resource allocation in real time; path planning ignores conflicts between resources and traffic congestion, which can easily lead to scheduling failures; the lack of resource availability prediction makes it impossible to identify risky resources in advance; the lack of a data feedback mechanism makes it difficult for the strategy to self-optimize; and it is difficult to achieve global optimization under multi-objective coordination, which limits the intelligence and practicality of the system.

[0007] Therefore, in order to address the above problems, there is an urgent need for a collaborative optimization method and system for logistics routes. Summary of the Invention

[0008] Technical problems to be solved

[0009] To address the shortcomings of existing technologies, this invention provides a logistics route collaborative optimization method and system, which solves the problem of delivery resource conflicts caused by the simultaneous operation of unmanned delivery vehicles and human delivery personnel in cities.

[0010] Technical solution

[0011] To achieve the above objectives, the present invention provides the following technical solution: a logistics path collaborative optimization method and system, comprising: S1, receiving user order tasks, collecting delivery data in real time and synchronizing delivery resource status, and performing real-time evaluation of task and resource allocation; S2, establishing a traffic map network, marking potential delivery conflict points, initially generating delivery paths, and performing real-time collaborative optimization of unmanned vehicles and delivery personnel paths based on path conflict risk assessment values; S3, continuously sensing the status of delivery resources, performing predictive analysis of the availability and execution capability of delivery resources, and implementing optimization measures based on the analysis results; S4, identifying emergencies, transferring and redistributing tasks, dynamically reallocating delivery resources and paths, and performing redistribution pressure assessment; S5, analyzing system log records, constructing a training dataset, evaluating scheduling effects and optimization strategies, and generating reports.

[0012] Furthermore, the specific process of receiving user order tasks, collecting delivery data in real time, and synchronizing delivery resource status is as follows: User order information is received through user application software and merchant system interfaces, including: start and end locations, delivery time limits, item type, order time, and customer preferences. Item types include fragile items, heavy items, and cold chain items. Delivery data is collected in real time, specifically: Real-time data on the status of unmanned vehicles and delivery personnel is reported through in-vehicle IoT terminal devices and delivery personnel mobile software, including: unmanned vehicle battery level, real-time location of delivery resources, real-time delivery speed, delivery load, and delivery resource availability. Energy consumption parameters of unmanned vehicle equipment are obtained through in-vehicle sensors. Delivery personnel weight data is collected, and based on the delivery personnel weight data and real-time delivery speed, the current delivery personnel's physical fitness index is obtained through a comprehensive analysis using the metabolic equivalent method. All unmanned vehicle and delivery personnel status data are synchronized in real time, all order information and delivery data are uploaded to the delivery database, and the total number of resources and tasks are analyzed and summarized in real time.

[0013] Furthermore, the specific process for real-time evaluation of task and resource allocation is as follows: Obtain the order start location and real-time location of delivery resources; use the map service interface to obtain the distance from the current resource i to the start location of task j; obtain the total delivery distance based on the distance from the current resource i to the start location of task j and the order termination location, and calculate the total delivery time based on the real-time delivery speed of the delivery resources; obtain the total delivery distance, delivery load, and equipment energy consumption parameters, and use linear regression to obtain the energy consumption of the unmanned vehicle resource i executing task j; obtain the energy consumption of the delivery person resource i executing task j based on the delivery person's weight data and the total delivery time using the metabolic equivalent method; calculate the remaining time to complete the current task based on the order start and termination locations, the real-time location of delivery resources, and the real-time delivery speed; obtain the user's expected delivery time from the customer preferences in the order information; subtract the user's expected delivery time from the sum of the total delivery time and the remaining time to complete the current task, and multiply by the timeout penalty weight factor; and then... Multiply the distance from source i to the starting position of task j by the distance weight factor; multiply the total delivery time by the time consumption weight factor; multiply the energy consumption of resource i executing task j by the energy consumption weight factor; sum these four products to obtain the total penalty term; iterate through all resource and task allocation combinations. If resource i is assigned to execute task j, the optimization decision variable is 1; if resource i is not assigned to execute task j, the optimization decision variable is 0; multiply the total penalty term by the optimization decision variable, and based on the total number of resources and the total number of tasks, accumulate the results of each pair of task and resource combinations to obtain the task resource allocation evaluation value; compare the task resource allocation evaluation value with the allocation threshold in real time. If the task resource allocation evaluation value is greater than or equal to the allocation threshold, the task reassignment mechanism is triggered to automatically find a better combination. If a resource continues to cause the value to exceed the allocation threshold, resource elimination and rate limiting are implemented, and the task is marked as a failure and recorded in the log. The allocation result is output to the allocation decision table and stored in the scheduling cache queue.

[0014] Furthermore, the specific process of establishing a traffic map network, marking potential delivery conflict points, and initially generating delivery routes is as follows: A basic road network structure is established based on the urban road network; traffic data is acquired by connecting to the urban traffic open platform and camera sensors; real-time weighting is performed to determine the congestion level and travel time of road segments; historical task trajectories are analyzed in real-time based on historical order information, delivery data, and historical allocation results; density clustering is used to identify high-frequency intersection areas between unmanned vehicles and delivery personnel, and areas with high probability of path overlap between unmanned vehicles and delivery personnel are marked; according to the basic road network structure, intersections are used as nodes and roads as edges; a graph neural network model is trained using historical task trajectories and traffic data; the starting and ending delivery positions are input, and the graph neural network model predicts the estimated value from each node to the target ending position; areas with high frequency of intersection and high probability of overlap between unmanned vehicles and delivery personnel are considered in real-time, and the path generation process is guided to prioritize adjacent nodes with better estimates; finally, a node sequence of a preliminarily optimized task path is output.

[0015] Furthermore, the specific process of real-time collaborative optimization of unmanned vehicle and delivery personnel paths based on path conflict risk assessment values ​​is as follows: Obtain the node sequence of the task paths output in the path generation step, and analyze the number of overlapping path points for task j and task k; based on the node sequences of task j and task k, take the path with the most nodes as the total number of path points for the two tasks; obtain the real-time delivery speeds of the unmanned vehicle and delivery personnel, and calculate the speed difference between the two delivery resources at the same location for task j and task k based on the speed difference at the point of overlap between the unmanned vehicle and delivery personnel according to the node sequence; based on the sliding time window length, calculate the average delivery speed by obtaining the average real-time delivery speed of all active delivery resources reported in real-time by the vehicle-mounted IoT terminal device interface and the delivery personnel's mobile software; obtain the road segment congestion level as the traffic density score for path point p; multiply the ratio of the number of overlapping path points for task j and task k to the total number of path points for the two tasks by the path overlap weight factor; multiply the speed difference between the two delivery resources at the same location by the average speed difference... The ratio of delivery speed is multiplied by the speed difference weighting factor; the traffic density score is multiplied by the traffic density weighting factor; the sum of these three products yields the path conflict risk assessment value; when the path conflict risk assessment value is less than the risk threshold, the path conflict risk is considered low, the delivery task is executed normally according to the current path, the path broadcast is maintained and the current path is synchronized to the terminal equipment, and the normal scheduling process begins; when the path conflict risk assessment value is greater than or equal to the risk threshold, the path buffer adjustment mechanism is triggered, conflict nodes and dense nodes in the path are identified, a buffer time window is set for each key node, allowing the path to slide and adjust within the buffer during execution, and when path conflict is unavoidable, a one-minute delay and pre-stop emergency measures are automatically introduced; based on the traffic map network, while avoiding the conflict nodes and dense nodes, adjacent nodes are reselected according to the estimated distance from each node to the target termination position, resulting in an adjusted task path node sequence; flow control and task transfer strategies are implemented for high-frequency conflict resources, and conflict point information is recorded in the high-risk node database.

[0016] Furthermore, the specific process of continuously sensing the status of delivery resources and predicting their availability and execution capabilities is as follows: acquire the status data of unmanned vehicles and delivery personnel, construct a status feature vector, and use it as the training set for the Long Short-Term Memory Network model. Extract, encode, and align key features of various data types, such as real-time location of delivery resources, battery level of unmanned vehicles, real-time delivery speed, and delivery load, and transform them into corresponding status feature vectors. Then, integrate them into a status feature vector and use the Long Short-Term Memory Network model to perform real-time status prediction, predict whether the service capability will decline within a certain period of time, and output a service availability score from 0 to 1 based on the status prediction.

[0017] Further, the specific process of implementing optimization measures based on the analysis results is as follows: Obtain the service availability score predicted by the Long Short-Term Memory network model to obtain the service availability score at the current moment; obtain the current unmanned vehicle status data to obtain the current unmanned vehicle battery level; obtain historical unmanned vehicle battery level and delivery personnel physical fitness index from the delivery database, select the lowest value, and use a quantile regression algorithm to fit and obtain the minimum acceptable battery level and physical fitness index; obtain the current delivery load, and use a logistic regression algorithm to fit and analyze the delivery load with the highest probability of successful delivery to obtain the maximum resource carrying capacity; multiply the current service availability score by the historical availability weight factor; subtract the ratio of the current unmanned vehicle battery level and delivery personnel physical fitness index to the minimum acceptable battery level and physical fitness index by a constant, and then multiply by the energy influence weight factor; use a constant... Subtract the ratio of the current delivery load to the maximum resource carrying capacity, and then multiply it by the load impact weighting factor; sum these three products to obtain the dynamic update value of resource availability after a certain period of time; compare the dynamic update value of resource availability with the availability threshold in real time; when the dynamic update value of resource availability is less than the availability threshold, it is considered a low availability state, the resource's order acceptance qualification is automatically suspended, the task reassignment mechanism is triggered, and the task is transferred to a high availability resource so that the low availability resource is no longer used as the delivery resource for the task; mark the low availability resource as pending maintenance, enter status monitoring, and record the dynamic update value of resource availability and task execution performance; when the dynamic update value of resource availability is greater than or equal to the availability threshold, it is considered a high availability state, continues to participate in task matching and path optimization, prioritizes matching light-load, short-distance tasks, and records the dynamic update value of resource availability.

[0018] Furthermore, the specific process for identifying emergencies, transferring and redistributing tasks, dynamically reallocating delivery resources and routes, and conducting pressure assessments for reallocation involves: real-time monitoring of abnormal indicators, including: unmanned vehicle battery level below the battery threshold, resources continuously stationary at non-task nodes exceeding the parking threshold, and the difference between the sum of the total delivery time and the time required to complete the current task and the customer's expected delivery time exceeding the risk threshold; real-time querying of all available resources within the city based on the vehicle's IoT terminal interface and the delivery person's mobile software; re-matching tasks for obstructed resources based on the task resource allocation assessment value, and then... Before matching, determine if the task violates service standards. If so, cancel the rematch and mark it as a failed task, recording it in the log. Obtain the current real-time location of resources and the starting location of the target task. Using the current real-time location of resources as the starting location of the path and the starting location of the target task as the target location of the path, predict the estimated value of each node to the target location of the path using a graph neural network model based on the basic road network structure, and prioritize the adjacent nodes with better estimates to generate a planned path. Record all rescheduling actions in the log. Obtain and count the number of task reschedulings recorded in the log to obtain the number of tasks that need to be rescheduled. Obtain resources. For highly available resources with availability dynamically updated values ​​greater than or equal to the availability threshold, the current number of available resources is calculated statistically. The number of task failures within the current time period is calculated from records marked as task failures in the scheduling log. The total number of scheduled tasks within the current time period is obtained by counting all order task allocation records within the period using the allocation decision table in the scheduling cache queue. The task redistribution pressure value is obtained by multiplying the ratio of the number of task failures to the total number of scheduled tasks within the current time period by a constant, and then by the ratio of the number of tasks requiring rescheduling to the current number of available resources. The task redistribution pressure value is compared in real-time with... Level 1 and Level 2 pressure thresholds: When the task redistribution pressure value is less than or equal to the Level 1 pressure threshold, it is considered a normal state, and task redistribution is executed according to the standard strategy; when the task redistribution pressure value is greater than the Level 1 pressure threshold but less than or equal to the Level 2 pressure threshold, it is considered a mild stress state, prioritizing avoidance of path conflict areas and extending the scheduling interval to alleviate pressure, and controlling the new order access rate; when the task redistribution pressure value is greater than the Level 2 pressure threshold, it is considered a high-pressure state, and the backup resource pool, task degradation mechanism, and regional rate limiting strategy are immediately activated, high-frequency failed tasks are marked as scheduling risks, and high-frequency resource calls automatically enter a cooling state.

[0019] Further, the specific process of analyzing system log records, constructing a training dataset, evaluating scheduling effectiveness and optimization strategies, and generating reports is as follows: Key log information is continuously collected, including: task matching logs, path planning call records, dynamic scheduling change records, task failure records, and their classification reasons; a high-quality training set is constructed based on the collected delivery data, and through labeling, multi-dimensional result labels are marked for whether task allocation was successful, whether paths conflicted, and whether resources were efficient; an ensemble regression model is used to predict task scheduling effectiveness, and a classification model is introduced to identify high-risk resources and task types; monthly strategy execution effectiveness reports are automatically generated, and multi-dimensional performance monitoring and strategy review are conducted by combining task completion rate, failure rate, and resource utilization rate indicators, with real-time feedback of evaluation results to the terminal.

[0020] Furthermore, it includes: a task receiving and allocation module, a path coordination and conflict avoidance module, a resource status perception and prediction module, a dynamic scheduling and replanning module, and a data learning and feedback module. The task receiving and allocation module receives user order tasks, collects delivery data in real time, synchronizes delivery resource status, and performs real-time evaluation of task and resource allocation. The path coordination and conflict avoidance module establishes a traffic map network, marks potential delivery conflict points, initially generates delivery routes, and optimizes the path coordination between unmanned vehicles and delivery personnel in real time based on path conflict risk assessment values. The resource status perception and prediction module continuously perceives the status of delivery resources, predicts and analyzes the availability and execution capability of delivery resources, and implements optimization measures based on the analysis results. The dynamic scheduling and replanning module identifies emergencies, performs task transfer and reallocation, dynamically reallocates delivery resources and routes, and assesses the pressure of reallocation. The data learning and feedback module analyzes system log records, constructs training datasets, evaluates scheduling effectiveness and optimization strategies, and generates reports.

[0021] Beneficial effects

[0022] The present invention has the following beneficial effects:

[0023] (1) This invention integrates the status data of unmanned vehicles and delivery personnel with order information, evaluates the allocation quality in real time through task resource allocation evaluation value, and introduces automatic threshold judgment and task reassignment mechanism, which significantly improves the accuracy and robustness of resource scheduling and effectively avoids resource idleness or overload.

[0024] (2) This invention generates paths by introducing graph neural networks on the basis of traditional road networks, marks conflict points by historical trajectories and traffic density data, and calculates path conflict risk assessment values ​​to dynamically optimize delivery routes. When the conflict risk exceeds the threshold, path buffering, pre-stop and path replanning mechanisms are automatically activated to decouple unmanned vehicles and delivery personnel in high-density scenarios, thereby improving traffic efficiency and safety.

[0025] (3) This invention constructs a state feature vector, combines it with a long short-term memory network model to predict future service availability, and uses power thresholds and load limits to fit the resource carrying capacity boundary, forming a dynamically updated resource availability value. Based on the prediction results, the scheduling participation of low-availability resources is proactively suspended, effectively avoiding task interruption and path waste, and improving the system scheduling foresight and stability.

[0026] (4) This invention collects multi-dimensional data such as task scheduling logs, path execution records, and failure classifications to construct a training dataset and introduces regression and classification models to predict and evaluate scheduling performance. Combined with the results of monthly strategy review reports, the system dynamically adjusts strategy parameters such as task allocation thresholds, path penalty factors, and resource priorities, forming a data-driven self-learning closed loop for scheduling strategies, which significantly enhances the system's environmental adaptability and self-optimization capabilities.

[0027] Of course, any product implementing this invention does not necessarily need to achieve all of the advantages described above at the same time. Attached Figure Description

[0028] Figure 1 This is a flowchart of a logistics route collaborative optimization method;

[0029] Figure 2 This is a structural diagram of a logistics route collaborative optimization system;

[0030] Figure 3 Assign evaluation values ​​to tasks corresponding to different resources. Detailed Implementation

[0031] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0032] Please see Figures 1-3This invention provides a technical solution: a logistics route collaborative optimization method and system, comprising: S1, receiving user order tasks, collecting delivery data in real time and synchronizing delivery resource status, and performing real-time evaluation of task and resource allocation; S2, establishing a traffic map network, marking potential delivery conflict points, initially generating delivery routes, and performing real-time collaborative optimization of unmanned vehicles and delivery personnel routes based on route conflict risk assessment values; S3, continuously sensing the status of delivery resources, performing predictive analysis of the availability and execution capability of delivery resources, and implementing optimization measures based on the analysis results; S4, identifying emergencies, transferring and redistributing tasks, dynamically reallocating delivery resources and routes, and performing redistribution pressure assessment; S5, analyzing system log records, constructing a training dataset, evaluating scheduling effects and optimization strategies, and generating reports.

[0033] Specifically, the process of receiving user order tasks, collecting delivery data in real time, and synchronizing delivery resource status is as follows: User order information is received through user application software and merchant system interfaces, including: start and end locations, delivery time limits, item type, order time, and customer preferences. Item types include fragile items, heavy items, and cold chain items; this information provides the basic input for subsequent task scheduling. Real-time delivery data collection specifically involves: reporting real-time data on the status of unmanned vehicles and delivery personnel through onboard IoT terminal devices and delivery personnel mobile software, including: unmanned vehicle battery level, real-time location of delivery resources, real-time delivery speed, delivery load, and delivery resource availability; this helps the system monitor resource health in real time and dynamically adjust task allocation strategies. Energy consumption parameters of unmanned vehicle equipment are obtained through onboard sensors; delivery personnel weight data is collected, and based on the delivery personnel weight data and real-time delivery speed, a metabolic equivalent method is used to comprehensively analyze and obtain the current delivery personnel's physical fitness index; this index is used to assess the continuous operating capacity of manual delivery resources and improve scheduling safety. Real-time synchronization of status data for all unmanned vehicles and delivery personnel; uploading all order information and delivery data to the delivery database; and real-time analysis and summarization of the total number of resources and tasks, providing key decision-making basis for subsequent resource matching and route generation.

[0034] In this implementation plan, by integrating user order information with real-time delivery resource data, dynamic evaluation and precise control of task allocation are achieved. The collection of status parameters such as unmanned vehicle battery level, delivery speed, and load improves the real-time performance and stability of resource scheduling; the analysis of delivery personnel's physical fitness index enhances the ability to judge the sustainability of human resources; and the system's synchronous aggregation of orders and resources provides reliable data support for subsequent route planning and task matching, thereby significantly improving overall scheduling efficiency and service reliability.

[0035] Specifically, the real-time evaluation process for task and resource allocation is as follows: First, the order's starting location and the real-time location of delivery resources are obtained. The distance from the current resource i to the starting location of task j is obtained using a map service interface. This distance, as a crucial component of scheduling costs, directly impacts the priority of resource selection. The total delivery distance is obtained based on the distance from the current resource i to the starting location of task j and the order's ending location. The total delivery time is then calculated based on the real-time delivery speed of the delivery resources. This total delivery time reflects timeliness requirements, ensuring tasks are completed within an acceptable timeframe. The total delivery distance, delivery load, and equipment energy consumption parameters are obtained, and linear regression is used to determine the energy consumption of autonomous vehicle resource i executing task j. Based on delivery personnel weight data and the total delivery time, the metabolic equivalent method is used to determine the energy consumption of delivery personnel resource i executing task j. Energy consumption estimation helps assess resource consumption levels and optimize cost control during scheduling. Based on the order's starting and ending locations, the remaining time to complete the current task is calculated using the real-time location of delivery resources and the real-time delivery speed. Customer preferences from the order information are used to obtain the user's expected delivery time. Introducing the expected time enhances the system's responsiveness to service quality. The total penalty term is calculated by subtracting the user's expected delivery time from the sum of the total delivery time and the time still needed to complete the current task, and then multiplying this sum by the timeout penalty weight factor. The distance from the current resource i to the starting position of task j is multiplied by the distance weight factor. The total delivery time is multiplied by the time consumption weight factor. The energy consumption of resource i in executing task j is multiplied by the energy consumption weight factor. The sum of these four products yields the total penalty term. This penalty term comprehensively reflects the cost level of resource matching and is a key reference for optimizing task allocation strategies. The algorithm iterates through all resource and task allocation combinations. If resource i is assigned to task j, the optimization decision variable is 1; otherwise, it is 0. The total penalty term is multiplied by the optimization decision variable, and based on the total number of resources and tasks, the results of each task and resource combination are accumulated to obtain the task resource allocation evaluation value. The task resource allocation evaluation value is compared with the allocation threshold in real time. If the evaluation value is greater than or equal to the threshold, a task reassignment mechanism is triggered to automatically find a better combination. This mechanism effectively reduces the resource misallocation rate and improves system response resilience. If a resource consistently exceeds the allocation threshold, resource eviction and rate limiting are implemented, and the task is marked as failed and recorded in the log. The allocation result is output to the allocation decision table and stored in the scheduling cache queue. This process helps to form a resource state evolution mechanism, providing a feedback basis for system self-learning and scheduling correction.

[0036] The specific formula for the task resource allocation evaluation value is as follows: ;

[0037] In the formula, Assign evaluation values ​​to task resources to assess the overall quality of a task resource allocation plan; For indexing delivery resources by number, Index the delivery tasks; This represents the total number of dispatchable delivery resources; The total number of delivery tasks to be assigned; To optimize the decision variable, we have , which represents whether to assign task j to resource i. If not assigned, the optimization decision variable is 0; if assigned, the optimization decision variable is 1. This represents the distance from the current resource i to the starting position of task j; Let i be the total delivery time for task j to be executed from its current location. Energy consumption for resource i to execute task j; It will take time for resource i to complete the current task; The user's expected delivery time for task j; The distance weighting factor is obtained by fitting the data through a linear regression algorithm based on the distance between delivery resources and delivery tasks and actual delivery cost data, and its range is between 0.3 and 2.0. The time consumption weighting factor is obtained through regression analysis based on the distance between delivery resources and delivery tasks, and the time required to complete the task, and ranges from 0.1 to 0.5. The energy consumption weighting factor is automatically learned using a Bayesian optimization algorithm based on actual delivery costs and energy consumption of delivery resources, and its range is between 0.005 and 0.01. The weighting factor for timeout penalties is determined by predicting the mapping between delay time and user churn based on total delivery time and expected delivery time data. This is obtained through logistic regression analysis and ranges from 0.5 to 5.0.

[0038] The distance weighting factor was set to 0.5, the time consumption weighting factor to 0.2, the energy consumption weighting factor to 0.01, and the timeout penalty weighting factor to 0.8. Under these weighting factors, different combinations of task resource allocation were performed. For different distances from resources to the task starting point, total delivery time, time remaining to complete the current task, energy consumption of resources to execute the current task, and user-expected delivery time, the task resource allocation evaluation value for different combinations of task resources was calculated. Table 1 shows the task resource allocation evaluation value data.

[0039] Table 1. Task Resource Allocation Evaluation Values

[0040] like Figure 3 As shown in Table 1, these are the task resource allocation evaluation values ​​for different resources corresponding to tasks provided in the embodiments of this application. Figure 3As can be seen, under the set weighting factors, the task resource allocation evaluation values ​​of the five groups of delivery resources when performing corresponding tasks under different conditions reflect the overall cost of task matching. The lower the task resource allocation evaluation value, the better the allocation scheme.

[0041] In this implementation plan, by integrating multiple factors such as distance, time consumption, energy consumption, and timeout risk, the system dynamically evaluates the quality of resource and task matching. This not only improves the accuracy and efficiency of task allocation but also enables automatic avoidance of unreasonable scheduling through a penalty mechanism. At the same time, the system implements flow control and elimination for abnormal resources and records failure information, providing data support for subsequent scheduling optimization and strategy self-learning, significantly enhancing the intelligence and adaptability of the scheduling system.

[0042] Specifically, the process of establishing a traffic map network, marking potential delivery conflict points, and initially generating delivery routes involves: establishing a basic road network structure based on the urban road network; accessing the urban traffic open platform and camera sensors to acquire traffic data; performing real-time weighting to determine the congestion level and travel time of road segments, providing a real-time and refined traffic perception foundation for route planning; and analyzing historical task trajectories in real time based on historical order information, delivery data, and historical allocation results, using density clustering to identify high-frequency intersection areas between unmanned vehicles and delivery personnel, and marking areas with a high probability of path overlap between unmanned vehicles and delivery personnel, which helps to identify potential path conflicts in advance and improve scheduling coordination. Regarding safety, based on the basic road network structure, intersections are used as nodes and roads as edges. A graph neural network model is trained using historical task trajectories and traffic data. The starting and ending positions of the delivery are input, and the graph neural network model predicts the estimated value from each node to the target ending position. This model can effectively capture the dynamic characteristics and traffic evolution patterns in the road network, and consider areas with high frequency of intersection and high probability of overlap between unmanned vehicles and delivery personnel in real time. It also guides the path generation process to prioritize the selection of adjacent nodes with better estimates, thereby reducing the possibility of path overlap and traffic conflicts, enhancing path executability and system throughput, and finally outputting a node sequence of a preliminary optimized task path.

[0043] In this implementation plan, by integrating real-time traffic perception and historical trajectory data, potential conflict areas in the delivery route are identified, and preliminary routes with low overlap and low congestion are intelligently generated using a graph neural network model. This not only improves the accuracy and robustness of route planning, but also effectively reduces the risk of conflicts between resources and enhances the system's scheduling and coordination capabilities in complex urban environments.

[0044] Specifically, the process of real-time collaborative optimization of unmanned vehicle and delivery personnel paths based on path conflict risk assessment values ​​is as follows: First, obtain the node sequence of the task paths output in the delivery path generation step, and analyze the number of overlapping path points between task j and task k. This overlap directly reflects the degree of sharing of path space between resources and is one of the core indicators for judging conflict risk. Then, based on the node sequences of task j and task k, take the path with the most nodes as the total number of path points for both tasks. This process ensures broad coverage during the assessment and improves the accuracy of risk identification. Next, obtain the real-time delivery speed of the unmanned vehicle and delivery personnel, and take the delivery speed at the point of overlap between the unmanned vehicle and delivery personnel based on the node sequence. The speed difference is the speed difference between two delivery resources in delivery tasks j and k at the same location; the larger the speed difference, the higher the probability of intersection waiting or obstruction. Based on the sliding time window length, the average delivery speed is calculated by obtaining the average real-time delivery speed of all active delivery resources reported in real time by the vehicle-mounted IoT terminal device interface and the deliveryman's mobile software. The road segment congestion level is obtained as the traffic density score of the path point p; the traffic density score can reflect the environmental congestion pressure of the path segment and is an environmental amplification factor for conflict risk. The ratio of the number of overlapping path points of tasks j and k to the total number of path points of the two tasks is multiplied by the path overlap weight factor. The speed difference between the two delivery resources at the same location is multiplied by the average delivery speed. The ratio of speeds is multiplied by the speed difference weighting factor; the traffic density score is multiplied by the traffic density weighting factor; the sum of these three products yields the path conflict risk assessment value. This assessment value integrates spatial overlap, speed interference, and environmental factors, and is an important basis for judging path safety and scheduling rationality. When the path conflict risk assessment value is less than the risk threshold, the path conflict risk is considered low, the delivery task is executed normally according to the current path, the path broadcast is maintained, and the current path is synchronized to the terminal equipment, entering the normal scheduling process. When the path conflict risk assessment value is greater than or equal to the risk threshold, the path buffer adjustment mechanism is triggered, identifying conflict nodes and dense nodes in the path, setting a buffer time window for each critical node, allowing... During path execution, sliding adjustments are made within the buffer zone. This mechanism reduces the incidence of instantaneous conflicts and improves resource flow. When path conflicts are unavoidable, a one-minute delay and pre-stop emergency measures are automatically introduced. Based on the traffic map network, while avoiding conflict nodes and dense nodes, adjacent nodes are reselected according to the estimated distance from each node to the target termination position, resulting in an adjusted task path node sequence. This helps to dynamically restore system scheduling stability and ensure the continuity of task execution. Flow control and task transfer strategies are implemented for high-frequency conflict resources, and conflict point information is recorded in the high-risk node database, providing data support for subsequent path planning and scheduling strategy training, and promoting the system's adaptive evolution and conflict early warning capabilities.

[0045] The specific formula for the path conflict risk assessment value is as follows: ;

[0046] In the formula, The path conflict risk assessment value for task j and task k at path point p represents the probability and severity of spatial overlap and operational interference between the selected paths of the unmanned vehicle and the deliveryman during the execution of the task. This represents the number of overlapping path points between task j and task k. The total number of waypoints for the two tasks; The speed difference between two delivery resources for delivery tasks j and k when they are in the same location; This represents the average delivery speed. Score the traffic density at point p; The path overlap weighting factor is based on a dataset of path point overlap and task failure rate. The relationship between path overlap rate and conflict probability is established using a logistic regression model and fitted by a Bayesian optimization algorithm. The value ranges from 0.3 to 0.6. The speed difference weighting factor is obtained by using a regression analysis algorithm based on the speed difference at the same location in real-time delivery tasks, and the nonlinear relationship between the speed difference and the task interruption time and congestion delay records generated during the actual delivery process. The range is between 0.1 and 0.4. The traffic density weighting factor is obtained by fitting traffic flow data, traffic congestion level, and route execution failure data through a logistic regression algorithm, and its range is between 0.2 and 0.5.

[0047] This implementation plan integrates three key factors—path overlap, speed difference, and traffic density—to assess delivery path conflict risks in real time and dynamically adjust path execution strategies based on the assessment results. This not only effectively reduces physical conflicts and scheduling interference between resources but also enhances path adaptability and traffic efficiency through buffering mechanisms and replanning schemes. Furthermore, it provides data support for subsequent strategy optimization and high-risk path identification, significantly improving the stability and intelligence level of multi-resource collaborative scheduling.

[0048] Specifically, the process of continuously sensing the status of delivery resources and predicting their availability and execution capabilities involves: acquiring status data of unmanned vehicles and delivery personnel, constructing status feature vectors as the training set for a long short-term memory network model, which effectively reflects the changing trends of resource operating status over time; extracting, encoding, and aligning key features from various data types, including real-time location of delivery resources, unmanned vehicle battery level, real-time delivery speed, and delivery load, to transform them into corresponding status feature vectors, which helps improve the model's perception accuracy and feature expression capabilities regarding resource status changes; integrating these into a status feature vector and using the long short-term memory network model for real-time status prediction, effectively capturing the evolution patterns of resource status in the short and long term, and predicting whether service capacity will decline over a period of time. This prediction provides a basis for the scheduling system to intervene in advance; and outputting a service availability score from 0 to 1 based on the status prediction, which can be used to dynamically adjust resource allocation strategies.

[0049] In this implementation plan, by constructing a feature vector that integrates multi-dimensional state information and using a long short-term memory network model to predict service capacity, a forward-looking assessment of the availability of delivery resources is achieved. This not only improves the scheduling system's ability to perceive and respond to changes in resource status, but also provides a refined scoring basis for task allocation, effectively ensuring the continuity and stability of resource scheduling.

[0050] Specifically, the process of implementing optimization measures based on the analysis results is as follows: First, obtain the service availability score predicted by the Long Short-Term Memory (LSTM) network model to obtain the service availability score at the current moment. This score reflects the overall service stability and execution capability of the resource in the current state. Second, obtain the current battery level of the unmanned vehicle based on its status data. Third, obtain historical battery levels of the unmanned vehicles and the physical fitness index of delivery personnel from the delivery database, select the lowest values, and use a quantile regression algorithm to fit and obtain the minimum acceptable battery level and physical fitness index of the resource. This is used to identify the safe operating boundary of the resource and avoid interruptions during the task. Fourth, obtain the current delivery load and, based on the historical delivery load obtained from the delivery database, use a logistic regression algorithm to fit and analyze the delivery load with the highest probability of successful delivery to obtain the maximum carrying capacity of the resource. This helps to dynamically adjust resource load limits and achieve intelligent load adaptation. Fifth, multiply the current service availability score by the historical availability weight factor. Sixth, subtract the ratio of the current unmanned vehicle battery level and delivery personnel physical fitness index to the minimum acceptable battery level and physical fitness index of the resource using a constant, and then multiply this by the energy influence weight factor to measure the degree of inhibition of battery level or physical fitness status on service capability. Finally, subtract the current delivery load by a constant. The ratio of the current load to the maximum carrying capacity of the resource is multiplied by the load impact weighting factor to identify whether the current load is close to the critical carrying capacity of the resource. The sum of these three products yields the dynamic update value of resource availability for a future period of time. This value comprehensively reflects the task execution stability and scheduling adaptability of the resource in the short term. The dynamic update value of resource availability is compared with the availability threshold in real time. When the dynamic update value of resource availability is less than the availability threshold, it is considered a low availability state. The resource is automatically suspended from accepting orders, triggering a task reassignment mechanism to transfer tasks to high availability resources. This prevents low availability resources from being used as task delivery resources, which helps improve task success rate and reduce the risk of abnormal system scheduling. Low availability resources are marked as pending maintenance and enter status monitoring. The dynamic update value of resource availability and task execution performance are recorded to provide feedback data for subsequent scheduling model optimization. When the dynamic update value of resource availability is greater than or equal to the availability threshold, it is considered a high availability state. It continues to participate in task matching and path optimization, prioritizing matching of light-load, short-distance tasks. The dynamic update value of resource availability is recorded to improve resource utilization efficiency and maintain the continuity and stability of task execution.

[0051] The specific formula for the dynamic update value of resource availability is as follows: ;

[0052] In the formula, For resources i in the future The resource availability is dynamically updated after a time period, with the result between 0 and 1, reflecting whether resources such as unmanned vehicles and delivery personnel have the ability to continuously complete delivery tasks in the future. Rate the service availability at the current moment; This refers to the current battery level of the driverless vehicle and the physical fitness index of the delivery personnel; The minimum acceptable power and physical fitness index for resources; Current delivery load; To maximize the carrying capacity of resources; The historical availability weighting factor is obtained by analyzing historical service availability scores, based on actual service capacity changes and status jump samples, and using an autoregressive model to analyze the autocorrelation of availability values ​​in the time series, with a range between 0.4 and 0.7. The energy impact weighting factor was obtained by analyzing the current battery level of the unmanned vehicle or the physical fitness index of the delivery personnel in history. Based on the statistical relationship between the remaining battery level and the probability of task interruption, the gradient boosting regression algorithm was used to fit the impact of energy on the availability score, with a range between 0.2 and 0.4. As a weighting factor for load impact, based on historical delivery load, the relationship between task completion time, failure rate and load ratio is analyzed, as well as the correspondence between load changes and resource speed decrease and power consumption. Load, speed, power consumption and interruption status are used as inputs, and the results are obtained by fitting using a multiple linear regression algorithm, with a range between 0.1 and 0.3.

[0053] In this implementation plan, by integrating service availability scores, power status, and load capacity, the future availability of resources is dynamically calculated, and scheduling strategies are automatically adjusted accordingly. This not only effectively predicts resource risks and avoids mid-process failures, but also achieves refined management of high and low availability states, improving task completion rate and the stability and intelligence of system scheduling.

[0054] Specifically, the process of identifying emergencies, transferring and redistributing tasks, dynamically reallocating delivery resources and routes, and assessing the pressure of reallocation involves: real-time monitoring of abnormal indicators, including: unmanned vehicle battery level below the battery threshold, resources remaining stationary beyond the parking threshold at non-task nodes, and the difference between the sum of the total delivery time and the time required to complete the current task and the customer's expected delivery time exceeding the risk threshold; these indicators effectively identify abnormal resource operation and potential task failure risks, serving as crucial criteria for triggering emergency dispatch mechanisms; real-time querying of all available resources within the city based on in-vehicle IoT terminal device interfaces and delivery personnel's mobile software reports; and assessment of hindered resources based on task resource allocation evaluation values. The source tasks are re-matched, and before re-matching, it is determined whether the task has violated the service standard. If so, the re-matching is canceled and the task is marked as failed and recorded in the log. This judgment mechanism can avoid unreasonable duplicate scheduling and improve system scheduling efficiency and task reliability. The current real-time location of the resource and the starting location of the target task are obtained. The current real-time location of the resource is used as the starting location of the path, and the starting location of the target task is used as the target location of the path. Based on the basic road network structure, a graph neural network model is used to predict the estimated value of each node to the target location of the path, and the adjacent nodes with better estimates are selected first to generate the planned path. This helps to ensure the feasibility and safety of alternative paths and improve task continuation efficiency. All re-matched tasks are re-matched. The behavior is recorded in the log for subsequent scheduling model training and system performance evaluation; the number of task rescheduling times recorded in the log is obtained and counted to determine the number of tasks that need to be rescheduled; highly available resources with dynamically updated availability values ​​greater than or equal to the availability threshold are obtained and their current availability is calculated; records marked as task failures in the scheduling log are counted to determine the number of task failures in the current time period; the total number of scheduled tasks in the current time period is obtained by counting the number of all order task allocation records in the allocation decision table of the scheduling cache queue; a constant is added to the ratio of the number of task failures in the current time period to the total number of scheduled tasks in the current time period, and then compared with the number of tasks that need to be rescheduled. The task redistribution pressure value is obtained by multiplying the number of tasks by the number of currently available resources. This value comprehensively reflects the current system task failure rate and resource supply and demand relationship, and is a key indicator for measuring the scheduling tension. The task redistribution pressure value is compared with the first-level and second-level pressure thresholds in real time. When the task redistribution pressure value is less than or equal to the first-level pressure threshold, it is judged as a normal state, and task redistribution is executed according to the standard strategy. When the task redistribution pressure value is greater than the first-level pressure threshold but less than or equal to the second-level pressure threshold, it is judged as a mild tension state. Path conflict areas are avoided first, and the scheduling interval is extended to alleviate the pressure and control the new order access rate. This strategy helps to maintain system stability and prevent task backlog when resources are scarce.When the task redistribution pressure exceeds the secondary pressure threshold, a high-pressure state is identified. The backup resource pool, task degradation mechanism, and regional rate limiting strategy are immediately activated. High-frequency failed tasks are marked as scheduling risks, and high-frequency resource calls automatically enter a cooling-off state to ensure system scheduling resilience and reduce the impact of continuous overload on service quality.

[0055] The specific formula for the task redistribution pressure value is as follows: ;

[0056] In the formula, The task reallocation stress value reflects the current operational stress of the scheduling system under the background of insufficient resources and task failure; This represents the number of tasks that currently need to be rescheduled. This represents the current amount of available resources. This represents the number of task failures within the current time period. This represents the total number of scheduled tasks within the current time period.

[0057] In this implementation plan, by monitoring abnormal indicators in real time, re-matching tasks and generating alternative paths, and dynamically determining the system scheduling status in conjunction with the task redistribution pressure value, not only is the resource response efficiency improved under emergencies, but the scheduling stability under high pressure is also ensured through rate limiting, task degradation and backup resource pool mechanisms. At the same time, the rescheduling behavior and task failure results are recorded, providing data support for subsequent system performance evaluation and strategy optimization.

[0058] Specifically, the process of analyzing system log records, constructing a training dataset, evaluating scheduling effectiveness and optimization strategies, and generating reports involves: continuously collecting key log information, including task matching logs, path planning call records, dynamic scheduling change records, task failure records, and their classification reasons. These logs serve as the core data source for scheduling behavior and results, comprehensively reflecting key nodes and abnormal states during system operation; constructing a high-quality training set based on collected delivery data, and labeling tasks with multi-dimensional results such as whether task allocation was successful, whether paths conflicted, and whether resources were efficient. This labeling system helps enhance the model's ability to identify different scheduling scenarios and improves the accuracy of prediction and classification; using an ensemble regression model to predict task scheduling effectiveness, enabling quantitative evaluation of key indicators such as resource matching efficiency and task completion quality; introducing a classification model to identify high-risk resources and task types, facilitating the system's early identification of potential problem resources and high-failure-probability tasks, enhancing the foresight of scheduling decisions; automatically generating monthly strategy execution effectiveness reports, combining task completion rate, failure rate, and resource utilization rate indicators for multi-dimensional performance monitoring and strategy review, helping operators regularly review the execution effectiveness of scheduling strategies, promptly identify and correct scheduling bottlenecks; and providing real-time feedback of evaluation results to the terminal.

[0059] In this implementation plan, by continuously collecting key log information and constructing a high-quality training set, and by using an integrated regression model and a classification model to evaluate the scheduling effect and identify high-risk resources, multi-dimensional performance monitoring and strategy review of indicators such as task completion rate, failure rate, and resource utilization rate are achieved. The evaluation results are fed back to the terminal in real time, which helps the scheduling system to achieve closed-loop self-optimization and improve scheduling intelligence and operational efficiency.

[0060] Reference Figure 2 As shown, the second aspect of this invention provides a logistics path collaborative optimization system, applied to the aforementioned logistics path collaborative optimization method, comprising: a task receiving and allocation module, a path collaboration and conflict avoidance module, a resource status perception and prediction module, a dynamic scheduling and replanning module, and a data learning and feedback module. The task receiving and allocation module receives user order tasks, collects delivery data in real time, synchronizes delivery resource status, and performs real-time evaluation of task and resource allocation. The path collaboration and conflict avoidance module establishes a traffic map network, marks potential delivery conflict points, initially generates delivery paths, and performs real-time path collaboration optimization between unmanned vehicles and delivery personnel based on path conflict risk assessment values. The resource status perception and prediction module continuously perceives the status of delivery resources, performs predictive analysis of the availability and execution capability of delivery resources, and implements optimization measures based on the analysis results. The dynamic scheduling and replanning module identifies emergencies, performs task transfer and reallocation, dynamically reallocates delivery resources and paths, and performs reallocation pressure assessment. The data learning and feedback module analyzes system log records, constructs a training dataset, evaluates scheduling effects and optimization strategies, and generates reports.

[0061] In this implementation plan, the collaboration of various modules enables real-time evaluation of task and resource allocation, collaborative optimization of delivery routes, dynamic prediction and adjustment of resource availability, task reallocation and stress assessment under emergencies, and continuous evaluation and feedback on scheduling effects and optimization strategies, thereby improving the system's intelligent scheduling capabilities, route execution stability and overall operating efficiency.

[0062] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0063] The preferred embodiments of the present invention disclosed above are merely illustrative of the invention. These preferred embodiments do not exhaustively describe all details, nor do they limit the invention to the specific implementations described. Clearly, many modifications and variations can be made based on the content of this specification. This specification selects and specifically describes these embodiments to better explain the principles and practical applications of the invention, thereby enabling those skilled in the art to better understand and utilize the invention. The invention is limited only by the claims and their full scope and equivalents.

Claims

1. A method for collaborative optimization of logistics routes, characterized in that, include: S1 receives user order tasks, collects delivery data in real time and synchronizes delivery resource status, and performs real-time evaluation of task and resource allocation. S2, establish a traffic map network, mark potential delivery conflict points, generate preliminary delivery routes, and optimize the routes of unmanned vehicles and delivery personnel in real time based on the route conflict risk assessment value; S3 continuously senses the status of delivery resources, performs predictive analysis on the availability and execution capability of delivery resources, and implements optimization measures based on the analysis results; S4 identifies emergencies, transfers and redistributes tasks, dynamically reallocates delivery resources and routes, and assesses redistribution pressure. The specific process of identifying emergencies, transferring and redistributing tasks, dynamically reallocating delivery resources and routes, and assessing the pressure of reallocation is as follows: Real-time monitoring of abnormal indicators includes: unmanned vehicle battery level below the battery threshold, resources continuously stationary at non-task nodes exceeding the parking threshold, and the difference between the sum of the total delivery time and the time required to complete the current task and the customer's expected delivery time exceeding the risk threshold; real-time reporting and querying of all available resources in the city based on the vehicle-mounted IoT terminal device interface and delivery personnel's mobile software; re-matching tasks for obstructed resources based on task resource allocation evaluation values, and determining whether the task has violated service standards before re-matching. If so, the re-matching is canceled and marked as a task failure, recorded in the log; obtaining the current real-time location of resources and the starting location of the target task, using the current real-time location of resources as the path starting location and the starting location of the target task as the path target location, predicting the estimated value of each node to the path target location using a graph neural network model based on the basic road network structure, and prioritizing the adjacent nodes with better estimates to generate a planned path; recording all rescheduling behaviors in the log; The task rescheduling count recorded in the log is used to obtain the number of tasks that need to be rescheduled. High-availability resources with dynamically updated availability values ​​greater than or equal to the availability threshold are obtained, and the number of currently available resources is calculated. Records marked as task failures in the scheduling log are counted to obtain the number of task failures within the current time period. The total number of scheduled tasks within the current time period is obtained by counting all order task allocation records within the period using the allocation decision table in the scheduling cache queue. The task rescheduling pressure value is obtained by multiplying the ratio of the number of task failures to the total number of scheduled tasks within the current time period by a constant, and then by the ratio of the number of tasks that need to be rescheduled to the number of currently available resources. The task redistribution pressure value is compared with the first and second level pressure thresholds in real time. When the task redistribution pressure value is less than or equal to the first level pressure threshold, it is judged as a normal state and the task redistribution is executed according to the standard strategy. When the task redistribution pressure value is greater than the first-level pressure threshold and less than or equal to the second-level pressure threshold, it is judged as a mild stress state. Priority is given to avoiding path conflict areas and extending the scheduling interval to alleviate pressure and control the new order access rate. When the task redistribution pressure value exceeds the secondary pressure threshold, it is determined to be a high-pressure state. The backup resource pool, task degradation mechanism and regional flow restriction strategy are immediately activated. High-frequency failure tasks are marked as scheduling risks, and high-frequency resource calls automatically enter the cooling state. S5 analyzes system log records, constructs a training dataset, evaluates scheduling effectiveness and optimization strategies, and generates reports.

2. The logistics route collaborative optimization method according to claim 1, characterized in that, The specific process of receiving user orders, collecting delivery data in real time, and synchronizing the status of delivery resources is as follows: The system receives user order information through user application software and merchant system interface, including: origin and destination locations, delivery time limit, item type, order time and customer preferences. Item types include fragile items, heavy items and cold chain items. Real-time delivery data collection specifically includes: reporting real-time status data of unmanned vehicles and delivery personnel through onboard IoT terminal devices and delivery personnel mobile applications, including: unmanned vehicle battery level, real-time location of delivery resources, real-time delivery speed, delivery load, and delivery resource availability; acquiring energy consumption parameters of unmanned vehicle equipment through onboard sensors; collecting delivery personnel weight data, and obtaining the current delivery personnel's physical fitness index based on the delivery personnel's weight data and real-time delivery speed using the metabolic equivalent method; synchronizing all unmanned vehicle and delivery personnel status data in real time, uploading all order information and delivery data to the delivery database, and analyzing and summarizing the total number of resources and tasks in real time.

3. The logistics route collaborative optimization method according to claim 1, characterized in that, The specific process for real-time evaluation of task and resource allocation is as follows: The system obtains the order's starting location and the real-time location of delivery resources. It uses a map service interface to obtain the distance from the current resource i to the starting location of task j. Based on the distance from the current resource i to the starting location of task j and the order's ending location, it calculates the total delivery distance and the total delivery time based on the real-time delivery speed of the delivery resources. It obtains the total delivery distance, delivery load, and equipment energy consumption parameters, and uses linear regression to obtain the energy consumption of the unmanned vehicle resource i executing task j. Based on the delivery person's weight data and the total delivery time, it uses the metabolic equivalent method to obtain the energy consumption of the delivery person resource i executing task j. Based on the order's starting and ending locations, and using the real-time location and real-time delivery speed of the delivery resources, it calculates the remaining time to complete the current task. It obtains the customer's expected delivery time from the order information based on customer preferences. It subtracts the user's expected delivery time from the sum of the total delivery time and the remaining time to complete the current task, and multiplies this by a timeout penalty weight factor. It multiplies the distance from the current resource i to the starting location of task j by a distance weight factor. It multiplies the total delivery time by a time consumption weight factor. Finally, it multiplies the energy consumption of resource i executing task j by an energy consumption weight factor. The total penalty term is obtained by summing the results of these four products; all resource and task allocation combinations are iterated through. If resource i is assigned to execute task j, the optimization decision variable is 1, and if resource i is not assigned to execute task j, the optimization decision variable is 0; the total penalty term is multiplied by the optimization decision variable, and based on the total number of resources and the total number of tasks, the results of each pair of task and resource combinations are accumulated one by one to obtain the task resource allocation evaluation value. The task resource allocation evaluation value is compared with the allocation threshold in real time. If the task resource allocation evaluation value is greater than or equal to the allocation threshold, the task reallocation mechanism is triggered to automatically find a better combination. If a certain resource continues to cause the allocation threshold to be higher, resource eviction and rate limiting are implemented, and the task is marked as a failure and recorded in the log. The allocation result is output to the allocation decision table and stored in the scheduling cache queue.

4. The logistics route collaborative optimization method according to claim 1, characterized in that, The specific process of establishing a traffic map network, marking potential delivery conflict points, and initially generating delivery routes is as follows: A basic road network structure is established based on the urban road network. Traffic data is obtained by connecting to the urban traffic open platform and camera sensors, and real-time weighting is performed to determine the congestion level and travel time of road segments. Based on historical order information, delivery data, and historical allocation results, the system analyzes historical task trajectories in real time. Density clustering is used to identify high-frequency intersection areas between unmanned vehicles and delivery personnel, and areas with high probability of path overlap between unmanned vehicles and delivery personnel are marked. According to the basic road network structure, intersections are used as nodes and roads are used as edges. A graph neural network model is trained using historical task trajectories and traffic data. The system takes into account the delivery start and end positions and predicts the estimated value from each node to the target end position. It considers areas with high frequency of intersection and high probability of overlap between unmanned vehicles and delivery personnel in real time and guides the path generation process to prioritize adjacent nodes with better estimates. Finally, a node sequence of a preliminary optimized task path is output.

5. The logistics route collaborative optimization method according to claim 1, characterized in that, The specific process of optimizing the route coordination between unmanned vehicles and delivery personnel in real time based on the route conflict risk assessment value is as follows: Obtain the node sequence of the task path output in the step of generating the delivery path, and analyze the number of overlapping path points of task j and task k; and take the path with the most nodes as the total number of path points of the two tasks according to the node sequence of task j and task k; obtain the real-time delivery speed of the unmanned vehicle and the delivery person, and take the speed difference at the point where the unmanned vehicle and the delivery person overlap according to the node sequence to obtain the speed difference of the two delivery resources of delivery task j and task k at the same location. Based on the sliding time window length, the average delivery speed is calculated by taking the average value of the real-time delivery speeds of all active delivery resources reported in real time by the vehicle-mounted IoT terminal device interface and the deliveryman's mobile software; the road segment congestion level is obtained as the traffic density score of path point p; the ratio of the number of overlapping path points between tasks j and k to the total number of path points for the two tasks is multiplied by the path overlap weight factor; the ratio of the speed difference between the two delivery resources at the same location to the average delivery speed is multiplied by the speed difference weight factor; and the traffic density score is multiplied by the traffic density weight factor. The summation of these three products yields the path conflict risk assessment value; When the risk assessment value of the route conflict is less than the risk threshold, the risk of route conflict is considered to be low. The delivery task is executed normally according to the current route, and the route broadcast is maintained and the current route is synchronized to the terminal device, and the normal scheduling process is entered. When the path conflict risk assessment value is greater than or equal to the risk threshold, the path buffer adjustment mechanism is triggered. Conflicting nodes and dense nodes in the path are identified, and a buffer time window is set for each key node, allowing for sliding adjustments within the buffer during path execution. When path conflicts are unavoidable, a one-minute delay and pre-stop emergency measures are automatically introduced. Based on the traffic map network, while avoiding the conflicting nodes and dense nodes, adjacent nodes are reselected according to the estimated distance from each node to the target termination position, resulting in an adjusted task path node sequence. Flow control and task transfer strategies are implemented for high-frequency conflict resources, and conflict point information is recorded in the high-risk node database.

6. The logistics route collaborative optimization method according to claim 1, characterized in that, The specific process of continuously sensing the status of delivery resources and performing predictive analysis on the availability and execution capability of delivery resources is as follows: The system acquires state data of unmanned vehicles and delivery personnel, constructs state feature vectors, and uses them as the training set for a long short-term memory network model. By extracting, encoding, and aligning key features of various data types, such as real-time location of delivery resources, battery level of unmanned vehicles, real-time delivery speed, and delivery load, the system transforms these features into corresponding state feature vectors. These vectors are then integrated into a single state feature vector, which is used to perform real-time state prediction using the long short-term memory network model. The system predicts whether service capacity will decline over a period of time and outputs a service availability score from 0 to 1 based on the state prediction.

7. The logistics route collaborative optimization method according to claim 1, characterized in that, The specific process for implementing optimization measures based on the analysis results is as follows: The service availability score at the current moment is obtained by obtaining the service availability score predicted by the Long Short-Term Memory network model; the current battery level of the autonomous vehicle is obtained by obtaining the autonomous vehicle status data; historical battery levels of autonomous vehicles and delivery personnel's physical fitness index are obtained from the delivery database, and the lowest values ​​are selected. A quantile regression algorithm is used to fit and obtain the minimum acceptable battery level and physical fitness index; the current delivery load is obtained, and historical delivery loads are obtained from the delivery database. A logistic regression algorithm is used to fit and analyze the delivery load with the highest probability of successful delivery, thus obtaining the maximum carrying capacity of the resource; the service availability score at the current moment is multiplied by the historical availability weight factor; the ratio of the current autonomous vehicle battery level and delivery personnel's physical fitness index to the minimum acceptable battery level and physical fitness index of the resource is subtracted by a constant, and then multiplied by the energy influence weight factor; the ratio of the current delivery load to the maximum carrying capacity of the resource is subtracted by a constant, and then multiplied by the load influence weight factor; Summing these three products yields the dynamically updated resource availability value for a future period of time. The system compares the dynamic update value of resource availability with the availability threshold in real time. When the dynamic update value of resource availability is less than the availability threshold, it is considered to be in a low availability state. The resource's order acceptance qualification is automatically suspended, and the task reassignment mechanism is triggered to transfer the task to a high availability resource so that the low availability resource is no longer used as the delivery resource for the task. The low availability resource is marked as pending maintenance and enters status monitoring to record the dynamic update value of resource availability and task execution performance. When the dynamic update value of resource availability is greater than or equal to the availability threshold, it indicates a high availability state and continues to participate in task matching and path optimization, prioritizing the matching of light-load, short-distance tasks, and recording the dynamic update value of resource availability.

8. The logistics route collaborative optimization method according to claim 1, characterized in that, The specific process of analyzing system log records, constructing a training dataset, evaluating scheduling effectiveness and optimization strategies, and generating reports is as follows: Continuously collect key log information, including: task matching logs, path planning call records, dynamic scheduling change records, task failure records and their classification reasons; build a high-quality training set based on the collected delivery data, and mark multi-dimensional result labels such as whether task allocation was successful, whether the path conflicted, and whether resources were efficient through labeling; use an ensemble regression model to predict task scheduling performance, and introduce a classification model to identify high-risk resources and task types; automatically generate a strategy execution performance report every month, and combine task completion rate, failure rate, and resource utilization rate indicators to conduct multi-dimensional performance monitoring and strategy review, and provide real-time feedback of evaluation results to the terminal.

9. A logistics route collaborative optimization system, employing a logistics route collaborative optimization method as described in any one of 1-8, characterized in that, include: The modules include: task reception and allocation, path coordination and conflict avoidance, resource status awareness and prediction, dynamic scheduling and replanning, and data learning and feedback. The task receiving and allocation module is used to receive user order tasks, collect delivery data in real time and synchronize delivery resource status, and perform real-time evaluation of task and resource allocation. The path coordination and conflict avoidance module is used to establish a traffic map network, mark potential delivery conflict points, initially generate delivery routes, and optimize the path coordination between unmanned vehicles and delivery personnel in real time based on the path conflict risk assessment value. The resource status perception and prediction module is used to continuously perceive the status of delivery resources, predict and analyze the availability and execution capability of delivery resources, and implement optimization measures based on the analysis results. The dynamic scheduling and replanning module is used to identify emergencies, transfer and redistribute tasks, dynamically reallocate delivery resources and routes, and assess the pressure of redistribution. The data learning and feedback module is used to analyze system log records, construct training datasets, evaluate scheduling effects and optimization strategies, and generate reports.

Images