A method for unmanned aerial vehicle delivery and pickup path planning for mountainous area agricultural products

Abstract

The application discloses a UAV delivery and pickup collaborative path planning method for mountainous area agricultural products, which comprises the following steps: gathering and distributing delivery and pickup orders; calculating a differentiated rural revitalization factor based on order types and seasons, and giving high priority to the pickup orders of agricultural products in the harvest season; participating in the clustering distribution of orders to hubs by using the factor; adopting an improved variable neighborhood descent algorithm which is fused with a delivery and pickup collaborative operator to perform path optimization, generating a main flight plan, and synchronously generating a static backup path for each flight section; and finally dynamically selecting between the main path and the backup path according to real-time risks during execution. The application improves operation economy through delivery and pickup collaborative planning, realizes accurate assistance to farmers through seasonal factors, and guarantees flight safety through preset double paths, effectively solving the technical problems of two-way delivery difficulty of UAV logistics in mountainous areas, poor timeliness of agricultural products uplink and high flight risk.

Description

Technical Field

[0001] This invention relates to the field of logistics distribution and route planning technology, and in particular to a method for collaborative route planning of drone delivery and pickup for agricultural products in mountainous areas, which is especially suitable for rural revitalization scenarios that need to handle both downhill delivery and uphill pickup of agricultural products. Background Technology

[0002] The complex terrain and weak infrastructure in mountainous areas, coupled with the high costs and poor timeliness of traditional ground logistics, severely restrict the upward movement of agricultural products. Drone logistics, with its advantages of vertical takeoff and landing and high mobility, is considered an effective means to overcome the bottlenecks in mountainous logistics, demonstrating significant potential in improving delivery efficiency and saving costs in remote areas. However, when applying drone logistics on a large scale and routinely to support the specific mission of moving agricultural products to mountainous areas, its path planning technology still faces two core challenges that urgently need to be addressed.

[0003] First, at the task mode level, the operational economics are mismatched with the characteristics of mountainous orders. Current research and practice mostly focus on optimizing one-way delivery tasks or exploring external collaboration between drones and other transportation vehicles. However, mountainous agricultural product orders are typically characterized by "low density and high dispersion." If drones only perform one-way pickup or are planned separately from downhill delivery tasks, it will lead to high return empty load rates, making it difficult to reduce the cost per flight and support a sustainable business model. Existing mountain drone path planning technologies, such as CN121764189A, improve path optimization capabilities in complex terrain through multi-constraint modeling and intelligent search algorithms, but their task models are essentially still unidirectional path optimization oriented towards a single task and a single origin and destination. Existing research does not pay enough attention to the path planning problem of achieving "integrated delivery and pickup" two-way collaboration within a single drone flight, and lacks systematic modeling and optimization methods. Second, at the scheduling strategy level, the intelligence of planning is out of touch with the inherent needs of agriculture. Agricultural product sales are highly seasonal and extremely time-sensitive, with significant differences in the urgency and economic value of transportation for different types of agricultural products and at different harvest times. Technical solutions in mountainous plant protection, such as CN121189590A, demonstrate an advanced approach to differentiated scheduling based on risk perception. However, their scheduling priority depends on the risk of terrain obstacles and operational efficiency, rather than the economic value and shelf-life of agricultural products. Most existing path planning technologies treat orders as static, homogeneous task points, primarily optimizing geometric distances or simple time windows, failing to establish an intelligent scheduling mechanism deeply integrated with the agricultural production cycle and the attributes of agricultural products. Although emerging technologies such as digital twins and artificial intelligence offer new ideas for dynamic environment modeling and decision-making, how to transform these technological capabilities into core algorithmic factors that drive logistics scheduling in mountainous areas and accurately respond to agricultural conditions remains a research gap. This makes it difficult for existing systems to automatically identify and prioritize the rapid transportation needs of high-value, perishable agricultural products during critical windows such as harvest seasons, thus weakening the actual effect of drone logistics in helping farmers increase their income.

[0004] Therefore, there is an urgent need for an innovative path planning method that can fundamentally realize the "delivery and pickup coordination" and "agricultural intelligence" of drone logistics in mountainous areas. This method can improve operational economy through systematic two-way task optimization and achieve precise scheduling based on dynamic perception of the multi-dimensional attributes of agricultural products, thereby providing an efficient, practical and safe drone logistics solution for the upward movement of agricultural products in mountainous areas. Summary of the Invention

[0005] To address the core contradictions of high operating costs and the disconnect between scheduling strategies and the inherent needs of agriculture caused by the difficulty in coordinating drone delivery and pickup in mountainous areas, this invention aims to provide a drone delivery and pickup coordinated path planning method for agricultural products in mountainous areas. By introducing rural revitalization factors to achieve intelligent scheduling of agricultural conditions, and by adopting delivery and pickup coordinated optimization and static dual-path pre-generation mechanisms, the method significantly improves the economic efficiency of drone operation, the timeliness of agricultural product delivery, and flight safety in complex mountainous environments.

[0006] To achieve the above objectives, the technical solution adopted by the present invention is as follows: A collaborative path planning method for drone delivery and pickup of agricultural products in mountainous areas includes the following steps: Step 1: The backend server system adopts a hierarchical parameter determination and optimization mechanism, which runs as a background process. Step 2: The cloud platform obtains the delivery order set D and the pickup order set P within the planning period, performs data cleaning and standardization on the orders, and extracts key attributes such as category, harvesting time, weight and location for agricultural product orders to form structured task data; Step 3: Based on the relationship between the current time and the harvest season of agricultural products, determine whether the system is running in regular mode or farmer assistance mode, and load the weight vector of the preset dynamic rural revitalization factor in the corresponding mode. Based on the key attributes of the order, calculate the dynamic rural revitalization factor for each order through a five-dimensional coupling model that integrates agricultural time series factor, quality and timeliness factor, economic value factor, service contribution factor and regional balance factor. The lower the factor value, the higher the delivery priority of the order. Step 4: Using township-level logistics hubs as distribution centers, calculate the comprehensive cost of each order to which it belongs to each hub. The comprehensive cost integrates geographical distance, estimated time penalty, and the dynamic rural revitalization factor. Under the constraints of capacity and time, allocate the order to the hub with the lowest comprehensive cost to form an initial task cluster. Step 5: For each hub's mission cluster, perform an improved variable neighborhood descent search that incorporates a dedicated domain operator set for delivery and pickup coordination to generate the master flight plan; optimize the cargo curve and mission sequence in a single flight, with the optimization process aiming to minimize the total cost, and the total cost function explicitly includes the energy consumption changes caused by dynamic load changes in flight segments. Step 6: While generating the main flight path for each flight segment, a static backup path is generated simultaneously. The generation method is to use a risk avoidance disturbance operator to prohibit preset static high-risk geographical grid nodes in the path search space, thereby planning a verifiable detour path. Step 7: Send the complete plan, including the main flight plan and backup routes for each segment, to the UAV flight control system; during flight, the UAV calculates the comprehensive risk value based on real-time perception data, and autonomously switches between the main route and static backup route according to preset thresholds, or performs an emergency return under extreme risk.

[0007] In step 1, the hierarchical parameter determination and optimization mechanism covers the determination and adjustment of the following three types of parameters: Calibration of basic physical and performance parameters: For parameters that are strongly correlated with the physical characteristics of the UAV platform and the operating environment, including load sensitivity coefficient, basic energy consumption coefficient per unit distance, and terrain difficulty coefficient, a combination of experimental data regression analysis and historical operation data fitting method is used to determine them; Optimization of multi-objective decision weights: For weight parameters that need to be weighed among multiple operational objectives, including the weights of each dimension in the rural revitalization factor, the weights of energy consumption, time and priority penalties in the objective function, and the weights of each risk factor in the comprehensive risk value, a strategy combining offline multi-objective optimization and online adaptive learning is adopted for determination and optimization. Dynamic setting of operation and decision thresholds: For various judgment thresholds, including risk level thresholds, task clustering thresholds, and order merging weight thresholds, a statistical learning and prediction model-driven method is used to set them.

[0008] In step 3, the dynamic rural revitalization factor The calculation was performed using a five-dimensional coupled model, which includes: Agricultural time series factors Based on the official harvest season time window for contracted agricultural products, priorities are dynamically adjusted within the harvest season to achieve a gradual scheduling strategy of "early-season priority, late-season regular scheduling." The calculation formula is as follows: Where o represents the order object and t represents the planning time point, The time-series base value represents the highest priority intensity at the start of the harvest season. and These represent the start and end times of the harvest season, respectively. Quality Time Factor Based on the preservation kinetics of different agricultural products and combined with the harvest time, the real-time quality residual degree is calculated to quantify the urgency of transportation. The calculation formula is as follows: in, Agricultural product harvesting time Total shelf life; Economic value factor Taking into account both product unit price and order aggregation scale, priority will be given to ensuring the transportation of high value-added products, and economies of scale will be encouraged. The specific formula is as follows: in It is the basic value of economic value. , Sensitivity coefficients for unit price and scale, respectively. Order unit price, The system presets a maximum unit price. Order weight, The threshold for rewards for group purchases; Service Contribution Factor By quantifying users' historical behavior, an incentive feedback mechanism is established to guide user behavior and reduce the overall operating costs of the system. The calculation is based on objective behavioral data extracted from the order management module and the user history database. The specific formula is as follows: in, Basic contribution; k=1,2,3 correspond to the three dimensions of performance credit, order aggregation, and network collaboration, respectively. Performance credit index Reflecting the reliability of users' historical behavior, through users' historical on-time fulfillment rate. ∈[0,1] means When user historical data is insufficient, the system default value of 0.8 is used; Order aggregation metric Encourage user behavior to increase drone loading rate, the expression is: , This is the sum of the weights of all orders placed by the user on that day. For single-unit threshold; network collaboration index Incentivize users to utilize the system's optimized logistics nodes This helps simplify the final path and reduce scheduling complexity; Regional balance factor To monitor and compensate for regional imbalances in logistics services, prevent remote or small villages from being chronically deprived of timely service, and ensure fairness in basic logistics services. ,in The average waiting time for all served orders in the village where the order is located on that day; It is the shortest average waiting time among all villages currently; This is the balance coefficient, with a default value of 0.1, which can be adjusted according to operational strategies. Therefore, dynamic rural revitalization factors Where o represents the order object; t represents the planning time point; Let i represent the dynamic weight coefficient of the i-th factor, satisfying ; Let represent the evaluation function of the i-th dimension, corresponding to the agricultural time series factor, quality timeliness factor, economic value factor, service contribution factor, and regional balance factor, respectively; ε(t) represents the real-time dynamic adjustment term, used to respond to sudden weather, policy instructions, or emergency events.

[0009] In step 4, the overall cost is... The calculation formula is: in, The weighting coefficients are used to balance distance costs, time urgency, and the priority of rural revitalization; For orders The three-dimensional Euclidean distance between destination v and hub H; This is a penalty for delivery time estimated based on distance and average speed.

[0010] The specific steps of step 5 are as follows: Step 5.1, using task clusters Corresponding township logistics hubs Using a heuristic approach, a path is generated with the starting and ending points as the reference points. Initial feasible routes between all delivery points and pickup points within the area Set algorithm parameters: maximum number of iterations without improvement. Current optimal path And calculate its total cost. ; Step 5.2: The algorithm enters the main iteration loop. In each iteration, the preset set of cooperative neighborhood operators is called sequentially. The operator in the current path Perturbation to generate new paths ; Step 5.3: Repeat the iterative process of step 5.2. If continuous... If no lower-cost feasible solution is found in any of the iterations, the algorithm is considered to have converged. Step 5.4: The algorithm terminates, and the final optimal path is output. As the main flight plan for this mission cluster, the plan specifies the UAV's access sequence, the expected arrival and departure times of each node, and the load change curve throughout the entire process. The set of collaborative operators for pickup and delivery in step 5.2 includes a pickup task insertion operator, a load balancing exchange operator, an agricultural product dedicated line generation operator, a time window elastic adjustment operator, and a task merging and splitting operator, as detailed below: Pickup task insertion operator (N1): Attempts to insert an unassigned pickup point. Insert two adjacent access points in the current path and between; Load balancing exchange operator (N2): For a sequence in the path where "pickup point is immediately followed by delivery point", the operator... → The operator attempts to swap. and The access order becomes Optimize the real-time load curve during flight; Agricultural Product Dedicated Line Generation Operator (N3): When the number of agricultural product pick-up points in a certain geographical area exceeds a threshold... When triggered, the operator clusters all eligible pickup points within the region and uses a heuristic algorithm to generate a locally optimal "agricultural product collection line" sub-path. This sub-path is then inserted into or replaces the corresponding part of the main path as a whole. Time window flexibility operator (N4): Identifies all access points in the current path that have "hard violations" or "soft violations" within the time window. For orders with soft violations, the flexibility adjustment operator adjusts the time window within the farmer's allowed time range. Within the framework, the planned visit times are fine-tuned, and the cost of the entire adjusted route is reassessed. This mechanism provides a larger space of feasible solutions for route optimization without affecting the core interests of farmers. Task merging and splitting operator (N5): For multiple small-weight delivery or pickup orders with similar time windows at the same customer node, where the total weight does not exceed [a certain value], [the operator can] merge and split these orders. Under the premise of reducing the number of takeoffs and landings, and targeting individual weights exceeding [a certain threshold], [these tasks are] merged into a single composite task node. For large pickup orders, break them down into multiple sub-orders and plan them into the continuous flight of the same or adjacent drone flights to ensure transportation feasibility.

[0011] The total cost function in step 5 The expression is: in, These are the weighting coefficients. For the total energy cost, consider the flight segment from node i to node j. Its energy consumption The model is as follows: in The basic energy consumption coefficient per unit distance is determined by the aerodynamic efficiency and motor efficiency of the UAV itself. This represents the three-dimensional terrain distance. The average load per flight segment, of which and These are the weights of the drone when it leaves node i and arrives at node j, respectively. The load sensitivity coefficient, based on UAV dynamics experimental data, represents the relationship between energy consumption and increasing load. The terrain difficulty coefficient, The difficulty of flying a drone under relevant terrain conditions is measured by comprehensively calculating the altitude change rate of the flight segment and the historical average wind speed data of the area. Total time penalty cost; It is the total priority cost. The priority penalty term is a component of the total priority cost, and its expression is: In the formula Rural revitalization factors for order o; This refers to the actual completion time of order o; This is the expected completion time for the order.

[0012] Step 6 involves using a risk avoidance perturbation operator to prohibit preset static high-risk geographical grid nodes from entering the path search space, thereby planning a verifiable detour route as the static backup route. The specific steps are as follows: Step 6.1: After the main flight plan is generated in step 5, the risk avoidance disturbance operator is immediately triggered. The operator queries the geographic information database to identify whether the path crosses or approaches a preset static high-risk area. This path is called Path A. Step 6.2: If a risk exists, in the path search for this flight segment, all grid nodes falling within the high-risk area are prohibited from use. While avoiding the constraint of the risk area, a new path is recalculated from... arrive The feasible path is designated as a static backup path, namely Path B, and the flight distance of Path B can be appropriately increased. Step 6.3: Bind the A and B paths of each flight segment as a "path pair" and integrate all the path pairs of the flight segments into a flight plan containing the complete main flight sequence and the coordinates of the A / B paths of each flight segment. Before the mission is executed, send the plan to the flight control system of the corresponding UAV. Step 6.4: Before the start of each flight segment and during flight, the UAV calculates the current comprehensive risk value based on sensor and meteorological data. ,like , If the risk threshold is low, then fly along path A; if , If the risk threshold is high and the current planned route is A, the flight will automatically switch to route B for that segment; if during flight... ,in If the risk threshold is reached, the mission will be immediately aborted, and the aircraft will autonomously plan and fly to the nearest safe alternate landing point (Z path). The comprehensive risk value in step 6.4 Including the following aspects: Weather risk It reflects the impact of wind speed and wind shear on flight stability and energy consumption, and its expression is: ,in For real-time wind speed, For gust wind speed, The maximum design wind speed for safe flight of drones, This is the wind shear penalty coefficient; Terrain proximity risk This reflects the degree of proximity of the drone to dangerous terrain, and its expression is: ,in This represents the three-dimensional distance from the drone's current location to the nearest steep ridge, high-voltage power line, or other dangerous terrain features. The characteristic attenuation distance; Real-time weather risk This value reflects macro weather conditions such as precipitation and visibility. It is a scale value mapped from real-time received aviation weather reports or meteorological radar data. Static area risk The risk value, derived from a preset risk map, reflects the inherent risk of historically high-risk areas. This value is a baseline risk value (0 to 1) read from the "Geographic Information and Risk Map Module" and bound to the grid where the drone is currently located. The formula for calculating the comprehensive risk value in step 6.4 is as follows: in , , , The weights of each risk factor.

[0013] Compared with the prior art, the beneficial effects of the present invention are: (1) This invention systematically models the delivery and pickup tasks and uses load balance exchange operators, task merging and splitting operators to intelligently integrate delivery and pickup tasks in a single flight, which significantly improves the return load rate of UAVs and the task density of a single flight, thereby effectively reducing the average cost of a single flight and laying the algorithmic foundation for the large-scale and normalized operation of UAV logistics in mountainous areas.

[0014] (2) This invention creatively introduces a dynamic rural revitalization factor as the core scheduling driver. This factor, through a multi-dimensional coupling model of agricultural time-series factors, quality and time-efficiency factors, and economic value factors, enables the system to automatically identify and prioritize the rapid transportation needs of high-value and perishable agricultural products during the harvest season at the planning source. This is equivalent to opening up an "intelligent green channel" for agricultural products, which can significantly shorten the time for agricultural products to leave the mountains and reduce losses, directly helping farmers increase their income and revitalize the countryside.

[0015] (3) This invention innovatively adopts a dual-path mechanism of "pre-generated main path (A) + static backup path (B)". During the offline planning stage, this mechanism uses a known high-precision risk map to synchronously generate a verifiable backup path for each flight segment that bypasses all preset static high-risk areas, thus forming a reliable risk avoidance plan. This provides key safety redundancy for UAVs to perform missions in weak communication environments and significantly improves the overall robustness of the system in complex terrain.

[0016] In summary, this invention adopts a layered collaborative architecture of "cloud-based preliminary planning + edge-based lightweight execution." The cloud leverages powerful computing capabilities to complete global static planning, including delivery and retrieval coordination, priority scheduling, and dual-path generation. After the planning scheme is distributed, edge computing units deployed in township logistics hubs drive the UAVs to execute the plan, handling only lightweight path switching decisions based on real-time perception. This architecture minimizes reliance on high-bandwidth, low-latency communication, perfectly adapting to the reality of weak communication infrastructure in mountainous areas. Simultaneously, it reduces the cost and energy consumption of edge computing hardware, making the entire system easier to engineer and deploy. Attached Figure Description

[0017] Figure 1 The flowchart below shows the overall process of the UAV delivery and retrieval collaborative path planning method provided in this embodiment of the invention.

[0018] Figure 2 A schematic diagram of the rural revitalization factor calculation model and scheduling priority provided in the embodiments of the present invention.

[0019] Figure 3 This is a detailed flowchart of the optimized delivery and retrieval collaborative path provided in an embodiment of the present invention.

[0020] Figure 4 This is a schematic diagram illustrating the generation of two static paths, A (primary) and B (backup), for a single flight segment in an embodiment of the present invention.

[0021] Figure 5 This is a logic diagram for path switching during the dynamic execution phase in an embodiment of the present invention. Detailed Implementation

[0022] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0023] It should be noted that the following detailed description is exemplary and intended to provide further illustration of the invention. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0024] It should be noted that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the exemplary embodiments of the invention. As used herein, unless the context clearly indicates otherwise, the singular form is intended to include the plural form as well. Furthermore, it should be understood that the terms “comprising” and “having”, and any variations thereof, are intended to cover a non-exclusive inclusion, for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but includes other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0025] Where there is no conflict, the embodiments and features in the embodiments of the present invention can be combined with each other.

[0026] This embodiment uses a mountainous county as an example to demonstrate the practical application of the method of the present invention. Figure 1 As shown, this invention provides a drone-based collaborative route planning method for agricultural product delivery in mountainous areas. Its execution follows a complete logical flow from order aggregation to dynamic execution. The overall system deployment includes a county-level operation center, township-level logistics hubs, and drone execution units. The county-level center deploys a static planning engine backend server, while each township hub is configured with an edge computing server to achieve distributed computing and rapid response. Specifically, it includes the following steps: Step 1: System parameter determination and adaptive optimization. To achieve adaptive operation and continuous optimization of the method of this invention, the background operating system adopts a hierarchical parameter determination and optimization mechanism. This mechanism runs as a background process and provides a dynamically updated parameter set for the core planning algorithm. The backend server of the county-level operation center is equipped with the static planning engine of this invention. Edge computing servers are deployed at logistics hubs in each township, responsible for local task scheduling and UAV control. The system initializes by loading the following basic data: high-precision 3D geographic information and a static risk map; a rural economic calendar database containing the harvest seasons and characteristics of relevant agricultural products within the service area; and a UAV performance parameter library including the maximum payload, cruising speed, and basic power consumption coefficient of each UAV. After system startup, the parameter determination and optimization mechanism begins running as a background process. This mechanism preloads the initial parameter set and dynamically updates relevant parameters based on historical operating data and real-time feedback during system execution, achieving adaptive operation and continuous optimization of the method.

[0027] Step 2: Order Aggregation and Preprocessing. Before the start of each planning cycle, the cloud platform aggregates all orders to be processed for the next service day, including delivery order set D and pickup order set P. The platform cleans and standardizes the raw order data, generating structured data records for each order, including basic information such as order type, weight, time window, and geographical location. For agricultural product pickup orders, it also associates and extracts key attributes such as the corresponding agricultural product category and estimated harvest time.

[0028] Step 3: Dynamic Rural Revitalization Factor Calculation. Based on the preprocessed order data, the system queries the rural economic calendar database to determine whether the current time falls within the "official harvest season time window" for relevant agricultural products within the service area. If it is outside the harvest season, the system enters "normal mode" and loads the normal mode parameter set. If it is within the harvest season, the system automatically switches to "farmer assistance mode" and loads the corresponding parameter set for the rural revitalization factor calculation model.

[0029] When the system enters the farmer assistance mode, such as Figure 2 As shown, the system utilizes the parameter set under this model, combined with the specific attributes of each order, to calculate a dynamic rural revitalization factor for each order through a five-dimensional coupling model. The model integrates evaluation functions across five dimensions to form a comprehensive priority index. The lower the value of this factor, the higher the order's delivery priority.

[0030] Agricultural time series factors Based on the official harvest season time window for contracted agricultural products, priorities are dynamically adjusted within the harvest season to achieve a gradual scheduling strategy of "early-season priority, late-season regular scheduling." The calculation formula is as follows: Where o represents the order object and t represents the planning time point, The time-series base value represents the highest priority intensity at the start of the harvest season. and These represent the start and end times of the harvest season, respectively.

[0031] Quality Time Factor Based on the preservation kinetics of different agricultural products and combined with the harvest time, the real-time quality residual is calculated to quantify the urgency of transportation. The calculation formula is as follows: in, Agricultural product harvesting time Total shelf life.

[0032] Economic value factor Taking into account both product unit price and order aggregation scale, priority will be given to ensuring the transportation of high value-added products, and economies of scale will be encouraged. The specific formula is as follows: in It is the basic value of economic value. , Sensitivity coefficients for unit price and scale, respectively. Order unit price, The system presets a maximum unit price. Order weight, The threshold for rewards for group purchases.

[0033] Service Contribution Factor This system quantifies historical user behavior and establishes an incentive feedback mechanism to guide user behavior and reduce overall system operating costs. The calculation is based on objective behavioral data extracted from the order management module and the user history database. The specific formula is as follows: in, Basic contribution; k=1,2,3 correspond to the three dimensions of performance credit, order aggregation, and network collaboration, respectively. Performance credit index Reflecting the reliability of users' historical behavior, through users' historical on-time fulfillment rate. ∈[0,1] means When user historical data is insufficient, the system default value of 0.8 is used. (Order aggregation metric) Encourage user behavior to increase drone loading rate, the expression is: , The sum of the weights of all orders placed by a user on that day, and the order aggregation threshold. Network collaboration indicators Incentivize users to utilize the system's optimized logistics nodes It helps simplify the end path and reduce scheduling complexity.

[0034] Regional balance factor To monitor and compensate for regional imbalances in logistics services, prevent remote or small villages from being unable to receive timely services for extended periods, and ensure fairness in basic logistics services. in The average waiting time for all served orders in the village where the order is located on that day; It is the shortest average waiting time among all villages currently; This is the balance coefficient, with a default value of 0.1, which can be adjusted according to operational strategies.

[0035] Dynamic Rural Revitalization Factors Where o represents the order object; t represents the planning time point; Let i represent the dynamic weight coefficient of the i-th factor, satisfying ; Let represent the evaluation function of the i-th dimension, corresponding to the agricultural time series factor, quality timeliness factor, economic value factor, service contribution factor, and regional balance factor, respectively; ε(t) represents the real-time dynamic adjustment term, used to respond to sudden weather, policy instructions, or emergency events.

[0036] Step 4: Based on clustering, task allocation is carried out with township-level logistics hubs as the allocation center. The comprehensive cost of each order to be assigned to each hub is calculated. The comprehensive cost integrates geographical distance, estimated time penalty and the dynamic rural revitalization factor. Under the constraints of capacity and time, the order is allocated to the hub with the minimum comprehensive cost to form an initial task cluster. Specifically: if township hubs H1 and H2 are used as distribution centers, for orders... Its destination Calculate their allocation to and The overall cost. The calculation formula is: in, The weighting coefficients are used to balance distance costs, time urgency, and the priority of rural revitalization; For orders The three-dimensional Euclidean distance between destination v and hub H; For delivery time penalties based on distance and average speed estimates, calculate separately. , ,like In satisfying Under the premise of the hub's total daily capacity constraints, orders will be placed Assign to the hub with the lowest overall cost According to this algorithm, all orders are allocated to the corresponding hubs, forming... Task Cluster and Task Cluster .

[0037] Step 5: Optimize the delivery and retrieval collaborative path. Based on the improved Variable Neighborhood Descent (IVND) algorithm, optimize the delivery and retrieval collaborative path. For each township hub's task cluster, perform the following steps: S5.1: Using the hub as the center, generate an initial flight path for all points to be visited within the task cluster, employing a simple heuristic method.

[0038] S5.2: Variable neighborhood descent main loop, setting the maximum number of iterations and the no-improvement termination threshold, enters the following loop optimization process: like Figure 3As shown, following a preset order or an adaptive strategy, the following five dedicated neighborhood operators are invoked sequentially to perturb the current solution and generate a new solution: Operator N1 (Pickup Task Insertion Operator): Randomly select an unassigned pickup point, attempt to insert it between two adjacent access points in the current path, and evaluate the feasibility and cost changes after insertion.

[0039] Operator N2 (load balance exchange operator): Scans the path, identifies the sequence of "pickup point followed by delivery point", and attempts to exchange the access order of the two to optimize the load curve throughout the flight and reduce energy consumption.

[0040] Operator N3 (Agricultural Product Dedicated Line Generation Operator): Detects a densely distributed cluster of pickup points in geospatial space, uses the Traveling Salesman Problem (TSP) heuristic algorithm to generate a locally optimal continuous access dedicated line, and replaces the corresponding scattered access points in the original path with this dedicated line.

[0041] Operator N4 (Time Window Flexible Adjustment Operator): Identifies access points in the path that violate the time window, fine-tunes their planned access times within the flexible time range allowed by the farmer, and reassesses the feasibility and cost of the entire path.

[0042] Operator N5 (Task Merging and Splitting Operator): For multiple low-weight orders at the same node, merge them into a composite task under sufficient load constraints; for overweight orders, split them into multiple sub-orders and plan them into continuous trips.

[0043] For each new solution generated by the operator, rigorous constraint checks are performed, including load constraints, time window constraints, and maximum range constraints. The total cost of the new solution is calculated. If the new solution is feasible and its total cost is lower than the current optimal solution, then the new solution is accepted as the current optimal solution, and the improvement is recorded. Total cost function. The expression is: in, These are the weighting coefficients; For the total energy cost, consider the flight segment from node i to node j. Its energy consumption The model is as follows: in The basic energy consumption coefficient per unit distance is determined by the aerodynamic efficiency and motor efficiency of the UAV itself. This represents the three-dimensional terrain distance. The average load per flight segment, of which and These are the weights of the drone when it leaves node i and arrives at node j, respectively. The load sensitivity coefficient, based on UAV dynamics experimental data, represents the relationship between energy consumption and increasing load. Terrain difficulty coefficient ( This measures the difficulty of flying a drone under relevant terrain conditions, based on the rate of altitude change of the flight segment. This was calculated based on a combination of historical average wind speed data for the region. Total time penalty cost; It is the total priority cost. The priority penalty term is a component of the total priority cost, and its expression is: In the formula Rural revitalization factors for order o. This refers to the actual completion time of order o. This is the expected completion time for the order.

[0044] S5.3: If any operator produces an improvement in this round, the algorithm will try again starting from the first operator; if no improvement is achieved after traversing all operators, the no-improvement count will be incremented. When the no-improvement count reaches a preset threshold, or the total number of iterations reaches the upper limit, the algorithm terminates and outputs the current optimal solution as the main flight plan for this task cluster. This plan specifies in detail the UAV's takeoff time, visit point sequence, estimated arrival / departure time for each point, flight distance for each segment, and estimated energy consumption, while satisfying all hard constraints and minimizing the total operating cost.

[0045] like Figure 4 , Figure 5 As shown, step 6, static backup path pre-generation, involves simultaneously generating a static backup path (Path B) for each segment of the flight plan while optimizing the primary flight plan. This aims to provide pre-defined risk avoidance options for the primary path (Path A). The specific steps are as follows: Step 6.1, for the flight segment After the main path (Path A) is generated, the risk avoidance disturbance operator is immediately triggered. The operator queries the geographic information database to identify whether Path A crosses or approaches a preset static high-risk area.

[0046] Step 6.2: If a risk exists, all grid nodes falling within the high-risk area are prohibited from use in the path search for this flight segment. Under the constraint of avoiding the risk area, a new path is recalculated from... arrive The feasible path is designated as a static backup path (Path B). The range of Path B can be appropriately increased.

[0047] Step 7: Dynamic Execution and Risk Response. For each flight segment, path A and path B are bound together as a "path pair." All path pairs are integrated and compiled into a flight plan containing the complete main flight sequence and the coordinates of each segment's A / B paths. This plan is then sent to the corresponding UAV's flight control system before mission execution.

[0048] Before the start of each flight segment and during the flight, the drone calculates the current comprehensive risk value based on sensor and meteorological data. .like , If the risk threshold is low, then fly along path A; if , If the risk threshold is high and the current planned route is A, the flight will automatically switch to route B for that segment; if during flight... ,in If the risk threshold is reached, the mission will be immediately aborted, and the aircraft will autonomously plan and fly to the nearest safe alternate landing point (Z path).

[0049] Through the detailed process described above, this invention achieves economic collaborative planning and dynamic safety avoidance for two-way delivery and retrieval tasks in complex mountainous environments. Those skilled in the art will understand that the parameters, values, and scenarios in the above embodiments are merely examples and can be adjusted according to actual application conditions. Any reasonable modifications and substitutions made based on the core concept of this invention fall within the scope of protection claimed by this invention.

[0050] This invention provides a system for implementing the above method, comprising: Order Management Module: Used to receive, store, and differentiate between delivery orders and pickup orders, and to extract the spatiotemporal attributes of orders; Rural Economic Calendar Database: Stores information on crop harvest seasons, major agricultural product categories, and preservation characteristics in the storage area; Static planning engine: This is the core algorithm execution module, used to run the delivery and retrieval cooperative path planning method and generate the main flight plan and static backup paths; Geographic Information and Risk Map Module: Provides high-precision 3D terrain data and maps marked with static high-risk areas; The planning and monitoring module compiles and distributes the planning scheme to the drone, receives the drone's status and risk perception information, and supervises the execution of path switching logic.

[0051] A device includes a memory: a computer program storing the aforementioned collaborative path planning method for drone delivery and retrieval of agricultural products in mountainous areas, which is a computer-readable device; Processor: Used to implement the aforementioned collaborative path planning method for drone delivery and retrieval of agricultural products in mountainous areas when executing the computer program.

[0052] A computer-readable storage medium storing a computer program that, when executed by a processor, enables the implementation of the aforementioned method for balancing safety and energy efficiency in unmanned aerial vehicle (UAV) path planning.

Claims

1. A method for collaborative path planning of drone delivery and pickup for agricultural products in mountainous areas, characterized in that, Includes the following steps: Step 1: The backend server system adopts a hierarchical parameter determination and optimization mechanism, which runs as a background process. Step 2: The cloud platform obtains the set of delivery orders D and the set of pickup orders P within the planning period, performs data cleaning and standardization on the orders, and extracts key attributes, including category, picking time, weight and location, for agricultural product orders to form structured task data; Step 3: Based on the relationship between the current time and the harvest season of agricultural products, determine whether the system is running in regular mode or farmer assistance mode, and load the weight vector of the preset dynamic rural revitalization factor in the corresponding mode. Based on the key attributes of the order, calculate the dynamic rural revitalization factor for each order through a five-dimensional coupling model that integrates agricultural time series factor, quality and timeliness factor, economic value factor, service contribution factor and regional balance factor. The lower the factor value, the higher the delivery priority of the order. Step 4: Using township-level logistics hubs as distribution centers, calculate the comprehensive cost of each order to which it belongs to each hub. The comprehensive cost integrates geographical distance, estimated time penalty, and the dynamic rural revitalization factor. Under the constraints of capacity and time, allocate the order to the hub with the lowest comprehensive cost to form an initial task cluster. Step 5: For each hub's mission cluster, perform an improved variable neighborhood descent search that incorporates delivery and pickup cooperative operators to generate the master flight plan; optimize the cargo curve and mission sequence in a single flight, with the optimization process aiming to minimize the total cost, and the total cost function explicitly includes the energy consumption changes caused by dynamic load changes in flight segments. Step 6: While generating the primary flight path for each flight segment, simultaneously generate a static backup path; Step 7: Send the complete plan, including the main flight plan and alternative routes for each segment, to the UAV flight control system; During flight, the drone calculates a comprehensive risk value based on real-time perception data and autonomously switches between the main path and static backup path according to preset thresholds, or performs an emergency return under extreme risk conditions.

2. The method for collaborative path planning of drone delivery and retrieval for agricultural products in mountainous areas according to claim 1, characterized in that, In step 1, the hierarchical parameter determination and optimization mechanism covers the determination and adjustment of the following three types of parameters: Calibration of basic physical and performance parameters: For parameters that are strongly correlated with the physical characteristics of the UAV platform and the operating environment, including load sensitivity coefficient, basic energy consumption coefficient per unit distance, and terrain difficulty coefficient, a combination of experimental data regression analysis and historical operation data fitting method is used to determine them; Optimization of multi-objective decision weights: For weight parameters that need to be weighed among multiple operational objectives, including the weights of each dimension in the rural revitalization factor, the weights of energy consumption, time and priority penalties in the objective function, and the weights of each risk factor in the comprehensive risk value, a strategy combining offline multi-objective optimization and online adaptive learning is adopted for determination and optimization. Dynamic setting of operation and decision thresholds: For various judgment thresholds, including risk level thresholds, task clustering thresholds, and order merging weight thresholds, a statistical learning and prediction model-driven method is used to set them.

3. The method for collaborative path planning of drone delivery and pickup for agricultural products in mountainous areas according to claim 1, characterized in that, In step 3, the dynamic rural revitalization factor The calculation was performed using a five-dimensional coupled model, which includes: Agricultural time series factors Based on the official harvest season time window for contracted agricultural products, priorities are dynamically adjusted within the harvest season to achieve a gradual scheduling strategy of "early-season priority, late-season regular scheduling." The calculation formula is as follows: Where o represents the order object and t represents the planning time point, The time-series base value represents the highest priority intensity at the start of the harvest season. and These represent the start and end times of the harvest season, respectively. Quality Time Factor Based on the preservation kinetics of different agricultural products and combined with the harvest time, the real-time quality residual degree is calculated to quantify the urgency of transportation. The calculation formula is as follows: in, Agricultural product harvesting time Total shelf life; Economic value factor Taking into account both product unit price and order aggregation scale, priority will be given to ensuring the transportation of high value-added products, and economies of scale will be encouraged. The specific formula is as follows: in It is the basic value of economic value. , Sensitivity coefficients for unit price and scale, respectively. Order unit price, The system presets a maximum unit price. Order weight, The threshold for rewards for group purchases; Service Contribution Factor By quantifying users' historical behavior, an incentive feedback mechanism is established to guide user behavior and reduce the overall operating costs of the system. The calculation is based on objective behavioral data extracted from the order management module and the user history database. The specific formula is as follows: in, Basic contribution; k=1,2,3 correspond to the three dimensions of performance credit, order aggregation, and network collaboration, respectively. Performance credit index Reflecting the reliability of users' historical behavior, through users' historical on-time fulfillment rate. ∈[0,1] means When user historical data is insufficient, the system default value of 0.8 is used; Order aggregation metric Encourage user behavior to increase drone loading rate, the expression is: , This is the sum of the weights of all orders placed by the user on that day. For single-unit threshold; network collaboration index Incentivize users to utilize the system's optimized logistics nodes This helps simplify the final path and reduce scheduling complexity; Regional balance factor To monitor and compensate for regional imbalances in logistics services, prevent remote or small villages from being chronically deprived of timely service, and ensure fairness in basic logistics services. ,in The average waiting time for all served orders in the village where the order is located on that day; It is the shortest average waiting time among all villages currently; This is the balance coefficient, with a default value of 0.1, which is adjusted according to the operational strategy. Therefore, dynamic rural revitalization factors Where o represents the order object; t represents the planning time point; Let i represent the dynamic weight coefficient of the i-th factor, satisfying ; Let represent the evaluation function of the i-th dimension, corresponding to the agricultural time series factor, quality timeliness factor, economic value factor, service contribution factor, and regional balance factor, respectively; ε(t) represents the real-time dynamic adjustment term, used to respond to sudden weather, policy instructions, or emergency events.

4. The method for collaborative path planning of drone delivery and retrieval for agricultural products in mountainous areas according to claim 1, characterized in that, In step 4, the overall cost is... The calculation formula is: in, These are weighting coefficients used to balance distance costs, time urgency, and the priority of rural revitalization; For orders The three-dimensional Euclidean distance between destination v and hub H; This is a penalty for delivery time estimated based on distance and average speed.

5. The method for collaborative path planning of drone delivery and pickup for agricultural products in mountainous areas according to claim 1, characterized in that, The specific steps of step 5 are as follows: Step 5.1, using task clusters Corresponding township logistics hubs Using a heuristic approach, a path is generated with the starting and ending points as the reference points. Initial feasible routes between all delivery points and pickup points within the area Set algorithm parameters: maximum number of iterations without improvement. Current optimal path And calculate its total cost. ; Step 5.2: The algorithm enters the main iteration loop. In each iteration, the preset set of cooperative neighborhood operators is called sequentially. The operator in the current path Perturbation to generate new paths ; Step 5.3: Repeat the iterative process of step 5.

2. If continuous... If no lower-cost feasible solution is found in any of the iterations, the algorithm is considered to have converged. Step 5.4: The algorithm terminates, and the final optimal path is output. As the main flight plan for this mission cluster, the plan specifies the UAV access sequence, the expected arrival and departure times of each node, and the load change curve throughout the entire process.

6. The method for collaborative path planning of drone delivery and retrieval for agricultural products in mountainous areas according to claim 5, characterized in that, The set of collaborative operators for pickup and delivery in step 5.2 includes a pickup task insertion operator, a load balancing exchange operator, an agricultural product dedicated line generation operator, a time window elastic adjustment operator, and a task merging and splitting operator, as detailed below: Pickup task insertion operator (N1): Attempts to insert an unassigned pickup point. Insert two adjacent access points in the current path and between; Load balancing exchange operator (N2): For a sequence in the path where "pickup point is immediately followed by delivery point", the operator... → The operator attempts to swap. and The access order becomes Optimize the real-time load curve during flight; Agricultural Product Dedicated Line Generation Operator (N3): When the number of agricultural product pick-up points in a certain geographical area exceeds a threshold... When triggered, the operator clusters all eligible pickup points within the region and uses a heuristic algorithm to generate a locally optimal "agricultural product collection line" sub-path. This sub-path is then inserted into or replaces the corresponding part of the main path as a whole. Time window flexibility operator (N4): Identifies all access points in the current path that have "hard violations" or "soft violations" within the time window. For orders with soft violations, within the flexible time range allowed by the farmer... Within the framework, the planned visit times are fine-tuned, and the cost of the entire adjusted route is reassessed. This mechanism provides a larger space of feasible solutions for route optimization without affecting the core interests of farmers. Task merging and splitting operator (N5): For multiple small-weight delivery or pickup orders with similar time windows at the same customer node, where the total weight does not exceed [a certain value], [the operator can] merge and split these orders. Under the premise of reducing the number of takeoffs and landings, and targeting individual weights exceeding [a certain threshold], [these tasks are] merged into a single composite task node. For large pickup orders, break them down into multiple sub-orders and plan them into the continuous flight of the same or adjacent drone flights to ensure transportation feasibility.

7. The method for collaborative path planning of drone delivery and retrieval for agricultural products in mountainous areas according to claim 6, characterized in that, The total cost function in step 5 The expression is: in, These are the weighting coefficients. For the total energy cost, consider the flight segment from node i to node j. Its energy consumption The model is as follows: in The basic energy consumption coefficient per unit distance is determined by the aerodynamic efficiency and motor efficiency of the UAV itself. This represents the three-dimensional terrain distance. The average load per flight segment, of which and These are the weights of the drone when it leaves node i and arrives at node j, respectively. The load sensitivity coefficient, based on UAV dynamics experimental data, represents the relationship between energy consumption and increasing load. The terrain difficulty coefficient, The difficulty of flying a drone under relevant terrain conditions is measured by comprehensively calculating the altitude change rate of the flight segment and the historical average wind speed data of the area. Total time penalty cost; It is the total priority cost. The priority penalty term is a component of the total priority cost, and its expression is: In the formula Rural revitalization factors for order o; This refers to the actual completion time of order o; This is the expected completion time for the order.

8. The method for collaborative path planning of drone delivery and retrieval for agricultural products in mountainous areas according to claim 1, characterized in that, Step 6 involves using a risk avoidance perturbation operator to prohibit preset static high-risk geographical grid nodes from entering the path search space, thereby planning a verifiable detour route as the static backup route. The specific steps are as follows: Step 6.1: After the main flight plan is generated in step 5, the risk avoidance disturbance operator is immediately triggered. The operator queries the geographic information database to identify whether the path crosses or approaches a preset static high-risk area. This path is called Path A. Step 6.2: If a risk exists, in the path search for this flight segment, all grid nodes falling within the high-risk area are prohibited from use. While avoiding the constraint of the risk area, a new path is recalculated from... arrive The feasible path is designated as a static backup path, namely Path B, and the flight distance of Path B can be appropriately increased. Step 6.3: Bind the A and B paths of each flight segment as a "path pair" and integrate all the path pairs of the flight segments into a flight plan containing the complete main flight sequence and the coordinates of the A / B paths of each flight segment. Before the mission is executed, send the plan to the flight control system of the corresponding UAV. Step 6.4: Before the start of each flight segment and during flight, the UAV calculates the current comprehensive risk value based on sensor and meteorological data. ,like , If the risk threshold is low, then fly along path A; if , If the risk threshold is high and the current planned route is A, the flight will automatically switch to route B for that segment; if during flight... ,in If the risk threshold is reached, the mission will be immediately aborted, and the aircraft will autonomously plan and fly to the nearest safe alternate landing point. The comprehensive risk value in step 6.4 Including the following aspects: Weather risk It reflects the impact of wind speed and wind shear on flight stability and energy consumption, and its expression is: ,in For real-time wind speed, For gust wind speed, The maximum design wind speed for safe flight of drones, This is the wind shear penalty coefficient; Terrain proximity risk This reflects the degree of proximity of the drone to dangerous terrain, and its expression is: ,in This represents the three-dimensional distance from the drone's current location to the nearest steep ridge, high-voltage power line, or other dangerous terrain features. The characteristic attenuation distance; Real-time weather risk This value reflects macro weather conditions such as precipitation and visibility. It is a scale value mapped from real-time received aviation weather reports or meteorological radar data. Static area risk The risk value, derived from a preset risk map, reflects the inherent risk of historically high-risk areas. This value is a baseline risk value read from the "Geographic Information and Risk Map Module" and bound to the grid where the drone is currently located. The formula for calculating the comprehensive risk value in step 6.4 is as follows: in , , , The weights of each risk factor.

9. A system, characterized in that, Implementing the drone delivery and pickup collaborative path planning method for agricultural products transported in mountainous areas as described in any one of claims 1 to 8, comprising: Order Management Module: Used to receive, store, and differentiate between delivery orders and pickup orders, and to extract the spatiotemporal attributes of orders; Rural Economic Calendar Database: Stores information on crop harvest seasons, major agricultural product categories, and preservation characteristics in the storage area; Static planning engine: This is the core algorithm execution module, used to run the delivery and retrieval cooperative path planning method and generate the main flight plan and static backup paths; Geographic Information and Risk Map Module: Provides high-precision 3D terrain data and maps marked with static high-risk areas; The planning and monitoring module compiles and distributes the planning scheme to the drone, receives the drone's status and risk perception information, and supervises the execution of path switching logic.

10. A device, characterized in that, Includes a memory: a computer program storing the aforementioned method for collaborative path planning of drone delivery and retrieval for agricultural products in mountainous areas, which is a computer-readable device; Processor: Used to implement the aforementioned collaborative path planning method for drone delivery and retrieval of agricultural products in mountainous areas when executing the computer program.

Images