Distributed plant unmanned aerial vehicle multi-node distribution path planning method and system

Abstract

This invention discloses a distributed multi-node delivery path planning method and system for unmanned aerial vehicles (UAVs) in a factory area. The method acquires the factory delivery environment, UAV operating parameters, and delivery task information; constructs a three-dimensional delivery space model including functional nodes, obstacles, no-fly zones, and passable flight areas; generates a set of delivery nodes to be visited based on the delivery task and configures task constraint parameters; calculates the path cost between nodes and constructs a cost matrix; determines the target node access sequence by combining task constraints and dynamic penalty weights; and performs segmented optimization on the three-dimensional trajectories between adjacent nodes, followed by smoothing, constraint re-verification, and sequential stitching of the segmented trajectories to obtain the target multi-node delivery path. This scheme improves the executability, security, and task scheduling adaptability of multi-node continuous delivery paths.

Description

Technical Field

[0001] This invention relates to the field of drone path planning technology, and in particular to a method and system for multi-node delivery path planning of drones in distributed factory areas. Background Technology

[0002] With the development of intelligent manufacturing, flexible production, and automated in-plant logistics, material delivery tasks within factories are increasingly characterized by multiple batches, multiple nodes, and high timeliness. Drones, with their advantages of maneuverability, minimal ground-based access restrictions, and convenient deployment, can be used for material transfer between storage areas, production workshops, quality inspection areas, assembly areas, and temporary docking points, thus providing a new technological means for flexible delivery within distributed factory areas.

[0003] When drones perform delivery tasks within factory areas, path planning is a crucial factor affecting delivery efficiency and flight safety. Existing drone path planning methods typically require generating flight paths that meet obstacle avoidance requirements based on the start point, destination, obstacle areas, no-fly zones, and flight constraints. Related methods usually focus on how to reduce path length, improve obstacle avoidance capabilities, or enhance algorithm convergence in complex environments, and can solve the path planning problem for a single delivery task to some extent.

[0004] However, in actual delivery scenarios within distributed factory areas, drones often do not simply complete a single flight from a single origin to a single destination. Instead, they need to continuously perform delivery tasks between multiple functional nodes within the factory. For example, a drone may need to sequentially visit warehousing nodes, production workshop nodes, quality inspection nodes, assembly nodes, temporary docking nodes, or charging nodes, completing operations such as picking up goods, delivering goods, quality inspection transfers, assembly transfers, transit buffering, or charging between different nodes. In this case, path planning not only needs to determine feasible trajectories between adjacent nodes but also the optimal order of access among multiple delivery nodes to be visited.

[0005] Existing point-to-point path planning methods typically focus on a single origin and destination, making them difficult to apply directly to factory delivery scenarios where multiple delivery task nodes are accessed consecutively. If independent path planning is performed only between each pair of nodes without considering the overall access order between delivery nodes, the drone may end up detouring between multiple factory nodes, increasing delivery distance and task execution costs.

[0006] Furthermore, delivery route planning within a distributed factory area needs to consider multiple task constraints and flight constraints simultaneously. For example, in the same delivery task, the pickup node should typically be visited before the delivery node; during multiple tasks, the payload status of a drone needs to meet the maximum payload limit; the drone's remaining endurance needs to meet the flight requirements between adjacent nodes; and some delivery tasks may also be subject to time windows and task priorities. Existing methods, lacking a unified model for these task constraints, are prone to generating geometrically feasible but impractical path schemes in the actual delivery process.

[0007] Meanwhile, factory environments typically contain spatially restricted areas such as buildings, production equipment, material storage areas, densely populated areas, hazardous material storage areas, and temporary control zones. When drones fly between adjacent delivery nodes, they not only need to avoid obstacle areas and no-fly zones, but also need to meet flight attitude constraints such as cruising altitude, maximum permissible yaw angle, and maximum permissible pitch angle. For discrete waypoints obtained through intelligent optimization algorithms, directly connecting the waypoints can easily lead to abrupt transitions and uneven local tracks; if only discrete tracks are smoothed, the smoothed continuous tracks may re-enter obstacle areas or no-fly zones due to curve deviation, or violate drone flight attitude constraints.

[0008] Therefore, how to comprehensively consider multiple functional nodes of the factory area, the set of delivery tasks, the path cost between nodes, the order of node access, the constraints of drone operation, obstacles and no-fly zones, and perform smoothing and constraint re-verification after generating segmented tracks to form a drone multi-node delivery path that meets the actual needs of factory area delivery has become an urgent technical problem to be solved. Summary of the Invention

[0009] The technical problem to be solved by this invention is to provide a method and system for multi-node delivery path planning of UAVs in distributed factory areas. This method addresses the issue that existing UAV path planning methods mainly focus on point-to-point trajectory planning from a single starting point to a single destination, making them difficult to apply to continuous access to multiple delivery task nodes within a distributed factory area. Furthermore, this invention solves the problem that existing methods lack unified processing for multiple factory area functional nodes, delivery task sets, path costs between nodes, target node access order, and smoothed trajectory constraint re-verification, resulting in delivery paths that fail to meet the multi-node delivery needs of distributed factory areas.

[0010] To address the aforementioned technical problems, this invention provides a distributed factory area drone multi-node delivery path planning method, comprising:

[0011] The system acquires information on the distribution environment of the distributed factory area, drone operating parameters, and delivery task information to construct a three-dimensional distribution space model of the distributed factory area. The three-dimensional distribution space model of the distributed factory area includes multiple functional nodes of the factory area, obstacle areas, no-fly zones, and passable flight areas.

[0012] Based on the delivery task information, a set of delivery nodes to be accessed is determined from the multiple factory area functional nodes, and task constraint parameters are configured for each delivery node to be accessed in the set of delivery nodes to be accessed.

[0013] Based on the distributed three-dimensional delivery space model of the factory area, the path cost between any two delivery nodes to be accessed in the set of delivery nodes to be accessed is calculated, and the path cost matrix between nodes is constructed according to the path cost between nodes.

[0014] Based on the path cost matrix between nodes and the task constraint parameters, an optimization solution method based on the improved artificial lemming algorithm is used to optimize the node access order of the set of delivery nodes to be visited, and determine the target node access sequence.

[0015] According to the target node access sequence, the optimization solution method based on the improved artificial lemming algorithm is used to optimize the three-dimensional flight path between two adjacent delivery nodes to be visited, and obtain multiple segmented delivery paths.

[0016] The multiple segmented delivery tracks are smoothed, constraint re-verified, and sequentially spliced ​​to generate a target multi-node delivery path that satisfies the target node access sequence and UAV flight constraints.

[0017] Output the target multi-node delivery path, wherein the target multi-node delivery path includes the node access order of the UAV and the segmented track information between adjacent delivery nodes to be visited.

[0018] Furthermore, the distributed three-dimensional delivery space model of the factory area includes the factory operation space, the set of factory functional nodes, the set of obstacle areas, the set of no-fly zones, and the area where drones can fly.

[0019] The factory workspace is referred to as:

[0020]

[0021] Where x, y, and z represent the spatial coordinates of the UAV in the three-dimensional coordinate system. , These represent the horizontal boundaries of the factory's delivery operation space. These represent the minimum and maximum altitudes at which drones are permitted to fly within the factory area, respectively.

[0022] The set of functional nodes in the factory area is represented as follows:

[0023]

[0024] in, Indicates the take-off and landing nodes of the drone. This represents the i-th functional node in the factory area. Each of the aforementioned factory area functional nodes includes at least node spatial coordinates, node type, and node service attributes; the node type includes at least one of the following: warehousing node, production workshop node, quality inspection node, assembly node, temporary parking node, and charging node;

[0025] The set of obstacle regions is represented as follows:

[0026]

[0027] in, This represents the obstacle area in the k-th factory area. The obstacle region is represented by a cuboid enclosure, a cylinder enclosure, a sphere enclosure, a polyhedron enclosure, or a combination thereof;

[0028] The set of no-fly zones is represented as follows:

[0029]

[0030] in, This represents the r-th no-fly zone. The no-fly zone includes at least one of the following: densely populated areas, hazardous materials storage areas, production control areas, and temporary control areas.

[0031] The area where the drone can fly is the area obtained by removing the set of obstacle areas and the set of no-fly zones from the factory's operating space, and is represented as follows:

[0032]

[0033] in, Indicates the area where drones are permitted to fly. This represents the set of three-dimensional factory work spaces defined by the spatial boundaries of the factory work space. Let m represent the k-th obstacle region, and m represent the number of obstacle regions. Let r represent the r-th no-fly zone, and q represent the number of no-fly zones; the flight area that the UAV can pass through is used to constrain the UAV's candidate waypoints, segmented delivery paths, and target multi-node delivery routes.

[0034] Furthermore, the delivery task information includes a set of delivery tasks:

[0035]

[0036] in, This represents the s-th delivery task. And the task Represented as:

[0037]

[0038] in, This represents the pickup node for the s-th delivery task. This indicates the delivery node corresponding to the s-th delivery task. This indicates the material load corresponding to the s-th delivery task. This indicates the task priority corresponding to the s-th delivery task, which is determined by both the task urgency and the material importance level. This represents the delivery time window corresponding to the s-th delivery task. This indicates the node service demand type corresponding to the s-th delivery task. The node service demand type includes at least one of the following: pickup, delivery, quality inspection and transfer, assembly and transfer, temporary parking, charging and replenishment, and transit buffer.

[0039] Based on the pickup and delivery nodes corresponding to each delivery task in the delivery task set R, an initial task node set is determined from the factory area functional node set, and an auxiliary node set is determined from the factory area functional node set based on the UAV operating parameters, node service attributes, and delivery task constraints.

[0040] The set of delivery nodes to be visited is generated based on the initial task node set, the auxiliary node set, and the drone take-off and landing nodes; wherein, the auxiliary node set includes at least one of charging nodes, temporary docking nodes, and transit buffer nodes, and the nodes in the auxiliary node set are determined as auxiliary nodes to participate in the current path planning based on endurance constraints, obstacle avoidance requirements, or task connection requirements.

[0041] Furthermore, the path cost matrix between nodes is expressed as:

[0042]

[0043] in, This represents the path cost matrix between nodes. This represents the nodes in the set of delivery nodes to be accessed. To node The path cost between nodes, This indicates the number of nodes in the set of delivery nodes to be accessed;

[0044] The path cost between nodes The cost is determined by a weighted average of the estimated flight distance between nodes, flight fluctuation cost, spatial avoidance cost, time window penalty cost, load constraint penalty cost, and endurance constraint penalty cost, and is expressed as follows:

[0045]

[0046] in, Represents a node To node The estimated flight distance, Represents a node To node The cost of flight fluctuations, Represents a node To node The cost of space avoidance, Represents a node To node Time window penalty cost Represents a node To node The penalty cost of load constraint Represents a node To node The penalty cost of limited battery life , , , , These are the weighting coefficients for the corresponding cost items;

[0047] The cost of space avoidance According to the node To node The safe distance relationship between the candidate flight path and the set of obstacle areas and the set of no-fly zones is determined; the time window penalty cost Load constraint penalty cost and the penalty cost of range constraints The determination is based on the delivery time window, the drone's payload status, and the drone's remaining battery life, respectively.

[0048] Furthermore, the target node access sequence is represented as follows:

[0049]

[0050] in, This represents the optimized target node access sequence. This represents the delivery node to be visited corresponding to the kth visit location of the drone, and M represents the length of the node visit sequence.

[0051] The target node access sequence is determined by minimizing the total delivery cost, which is determined based on the path cost between nodes, pickup and delivery order constraints, load constraints, time window constraints, and task priority constraints. Its optimization objective is expressed as:

[0052]

[0053] in, This represents the candidate node access sequence. This represents the path cost between two adjacent nodes in the candidate node visit sequence. This indicates a penalty for incorrect pickup / delivery order. This indicates the load constraint penalty term. This indicates the time window constraint penalty term. This indicates a task priority penalty item. , , , These represent the weight coefficients of the corresponding penalty items. The dynamic penalty weight coefficients are determined based on at least one of the following: the urgency of the delivery task, the importance level of the materials, the type of delivery demand, the remaining margin of the delivery time window, the current load status of the drone, and the degree of constraint violation of the candidate node access sequence.

[0054] The dynamic penalty weight coefficient can be determined based on the base weight and the task status adjustment factor:

[0055]

[0056] Where m represents any penalty item among the following: delivery order, load capacity, time window, or task priority. This represents the base weight of the m-th type of penalty term. This represents the adjustment coefficient for the m-th type of penalty term. This represents the task state adjustment factor corresponding to the candidate node access sequence π.

[0057] The pickup and delivery order constraint is used to restrict the pickup node from being accessed before the delivery node in the same delivery task. The load constraint is used to restrict the load of the drone during the node access process from not exceeding the maximum allowable load. The time window constraint is used to restrict the time of the drone to arrive at or serve the corresponding node to meet the time requirements of the delivery task. The task priority constraint is used to enable the node corresponding to the higher priority delivery task to obtain a higher access priority in the node access order optimization.

[0058] By using the aforementioned dynamic penalty weight coefficients, the influence of pickup and delivery order constraints, load constraints, time window constraints, and task priority constraints on the total delivery cost can be adaptively adjusted under different delivery task states. This ensures that the target node access sequence not only has a lower path cost but also prioritizes urgent tasks, important material tasks, and delivery tasks with strong time window constraints.

[0059] Furthermore, the optimization solution method based on the improved artificial lemming algorithm is used to optimize the solution of the target node access sequence and the three-dimensional flight path between the two adjacent delivery nodes to be visited; the improved artificial lemming algorithm includes:

[0060] An improved chaotic mapping method is used to initialize the lemming population to generate candidate node access sequences for node access order optimization and candidate waypoint sequences for segmented path optimization.

[0061] The current global optimal individual is determined based on the fitness function value corresponding to the candidate solution. When the improved artificial lemming algorithm is used for node access order optimization, the fitness function is determined based on the total delivery cost with the introduced dynamic penalty weight coefficient. When the improved artificial lemming algorithm is used for segmented trajectory optimization, the fitness function is determined based on the segmented trajectory length, flight fluctuation cost, and trajectory constraint penalty cost.

[0062] During the iterative search process, the lemming individual is controlled to perform global exploration behavior or local development behavior according to the stage discrimination mechanism of the artificial lemming algorithm, and a physical strength judgment mechanism is introduced in the local development behavior to adjust the local search strategy according to the physical strength of the lemming individual.

[0063] After the lemming individuals complete their position updates, a cross-sectional optimization strategy is introduced to perform secondary optimization on the updated candidate solutions, and the candidate solutions before and after optimization are selected based on the fitness function value.

[0064] When the preset stopping condition is met, the current globally optimal individual is output as the target node access sequence and the optimized result of the segmented delivery trajectory.

[0065] Furthermore, the improved chaotic mapping method is an improved Tent chaotic mapping that introduces a random perturbation term, and the improved Tent chaotic mapping is expressed as:

[0066]

[0067] Where k represents the number of mapping iterations, This represents the chaotic sequence value obtained by the k-th mapping, and rand(0,1) represents a random number between [0,1]. This represents the random disturbance intensity coefficient, and normalizes chaotic sequence values ​​that exceed the [0,1] interval.

[0068] Mapping the chaotic sequence values ​​generated by the improved Tent chaotic mapping to the search space yields the initial positions of individual lemmings:

[0069]

[0070] in, This represents the initial position of the i-th lemming individual. Let represent the chaotic sequence value used to generate the i-th lemming individual, and lb and ub represent the lower bound and upper bound of the search space, respectively;

[0071] When the improved artificial lemming algorithm is used for node access order optimization, the initial position of the lemming individual is used as a random key vector, and a candidate node access sequence is generated based on the sorting result of the random key vector; when the improved artificial lemming algorithm is used for segmented trajectory optimization, the initial position of the lemming individual is mapped to a candidate trajectory point sequence.

[0072] Furthermore, the physical strength assessment mechanism includes:

[0073] After the improved artificial lemming algorithm enters local development behavior, physical parameters are generated for the current lemming individual:

[0074]

[0075] in, Let represent the physical strength parameters of the i-th lemming individual. Represents a random number within the interval [0,1]. and These represent the physical fitness parameter adjustment coefficients;

[0076] The physical parameters Compared with the preset physical strength threshold The comparison is performed, and the local search strategy for the current lemming individual is determined based on the comparison results;

[0077] Among them, when When, control the current lemming individual to execute a local foraging search strategy; when At that time, control the current lemming individual to execute a stamina-restoring search strategy;

[0078] According to the physical strength parameters Determine local search weights:

[0079]

[0080] in, This represents the local search weight of the i-th lemming individual. This represents the weighting adjustment coefficient. Represents the optimal physical state parameters. Indicates parameters related to physical strength fluctuations;

[0081] Based on the local search weight Adjust the search step size and perturbation intensity in the local foraging search strategy or the physical recovery search strategy.

[0082] Furthermore, the cross-sectional optimization strategy includes lateral cross-sectional operations and longitudinal cross-sectional operations;

[0083] The lateral crossover operation is used to reorganize information between corresponding dimensions of different lemming individuals. For the values ​​of the i-th and j-th lemming individuals in the d-th dimension, the lateral crossover candidate solution is represented as:

[0084]

[0085]

[0086] in, and Let represent the values ​​of the i-th and j-th lemming individuals in the d-th dimension, respectively. and Let represent the values ​​of the candidate solutions generated after lateral crossover in the d-th dimension. and A random number within the interval [0,1]. and This is the cross disturbance coefficient;

[0087] The vertical crossover operation is used to perturb information between different dimensions of the same lemming individual, for two different dimensions in the i-th lemming individual. and The vertical cross candidate solution is represented as:

[0088]

[0089] Fitness evaluation is performed on the lateral and longitudinal crossover candidate solutions, and individuals are selected for retention from the pre-crossover lemming individuals, lateral crossover candidate solutions, and longitudinal crossover candidate solutions based on the fitness function value. Specifically, when the improved artificial lemming algorithm is used for node access order optimization, node sequence validity correction is performed on the lateral and longitudinal crossover candidate solutions; when the improved artificial lemming algorithm is used for segmented trajectory optimization, trajectory constraint validity correction is performed on the lateral and longitudinal crossover candidate solutions.

[0090] Furthermore, the smoothing, constraint re-verification, and sequential splicing of multiple segmented delivery tracks includes:

[0091] Based on the target node access sequence, the segmented delivery track between two adjacent delivery nodes to be accessed is taken as the track segment to be smoothed.

[0092] The track segment to be smoothed is subjected to curve smoothing processing to obtain a smoothed continuous segmented track;

[0093] Discrete sampling is performed on the smoothed continuous segmented trajectory, and constraint re-verification is performed on the trajectory segments formed by the sampling points and adjacent sampling points based on the UAV's passable flight area, obstacle area, no-fly zone and UAV flight attitude constraints.

[0094] If the smoothed continuous segmented track does not meet the constraint re-verification, then auxiliary track points are inserted into the local track segments that do not meet the constraints, and the curve smoothing and constraint re-verification are performed again based on the track point sequence after the insertion of auxiliary track points.

[0095] The continuous segmented tracks that satisfy the constraint re-verification are sequentially spliced ​​together according to the target node access sequence to generate the target multi-node delivery path.

[0096] The UAV flight attitude constraints include at least one of cruising altitude constraints, maximum permissible yaw angle constraints, and maximum permissible pitch angle constraints; the curve smoothing process is implemented using B-spline curves, Bézier curves, polynomial curves, or combinations thereof. This invention also provides a distributed factory UAV multi-node delivery path planning system for executing the above-mentioned distributed factory UAV multi-node delivery path planning method, the system comprising:

[0097] The environmental modeling module is used to acquire information on the distribution environment of the distributed factory area, the operating parameters of the drones, and the delivery task information, and to construct a three-dimensional delivery space model of the distributed factory area.

[0098] The task node generation module is used to determine the set of delivery nodes to be accessed from multiple functional nodes in the distributed three-dimensional delivery space model of the factory area based on the delivery task information, and to configure task constraint parameters for each delivery node to be accessed in the set of delivery nodes to be accessed.

[0099] The inter-node cost matrix construction module is used to calculate the inter-node path cost between any two delivery nodes in the set of delivery nodes to be visited, and to construct the inter-node path cost matrix based on the inter-node path cost.

[0100] The node access sequence optimization module is used to optimize the node access sequence of the set of delivery nodes to be visited based on the path cost matrix between nodes and the task constraint parameters, and to determine the target node access sequence for the UAV to visit multiple delivery nodes in sequence; the node access sequence optimization module is also used to determine the dynamic penalty weight coefficient based on the urgency of the delivery task, the importance level of the materials, the type of delivery demand, the remaining margin of the delivery time window, the current load status of the UAV and the degree of constraint violation of the candidate node access sequence, and to determine the target node access sequence based on the dynamic penalty weight coefficient;

[0101] An improved artificial lemming algorithm solution module is used to optimize the solution of the target node access sequence and the three-dimensional flight path between two adjacent delivery nodes to be visited based on the improved artificial lemming algorithm;

[0102] The segmented trajectory planning module is used to plan the three-dimensional flight trajectory between two adjacent delivery nodes to be visited according to the target node access sequence, so as to obtain multiple segmented delivery trajectories.

[0103] The path processing module is used to smooth, re-verify, and sequentially splice the multiple segmented delivery tracks to generate a target multi-node delivery path.

[0104] The path output module is used to output the target multi-node delivery path.

[0105] The present invention also provides an electronic device, including a processor, a memory, and a computer program stored in the memory and capable of running on the processor, wherein the processor executes the computer program to implement the above-described UAV multi-node delivery path planning method for distributed factory areas.

[0106] The present invention also provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the above-described UAV multi-node delivery path planning method for distributed factory areas.

[0107] Compared with the prior art, the present invention has at least the following beneficial effects:

[0108] First, this invention integrates pickup nodes, delivery nodes, and auxiliary nodes such as charging nodes, temporary docking nodes, and transit buffer nodes within a distributed factory area into a set of delivery nodes to be accessed. This makes it suitable for path planning scenarios where drones continuously access multiple delivery nodes, avoiding the problem that existing point-to-point path planning methods are difficult to adapt to multi-node continuous delivery tasks.

[0109] Second, this invention constructs a path cost matrix between nodes based on a distributed three-dimensional delivery space model of the factory area, and comprehensively considers cost factors such as flight distance, flight fluctuations, spatial avoidance, time window, load and endurance, which can provide a unified cost basis for optimizing the access order between multiple delivery nodes to be accessed.

[0110] Third, by incorporating constraints such as pickup and delivery order, load, time window, and task priority during the node access order optimization process, this invention enables the generated target node access sequence to better meet the execution requirements of actual factory area delivery tasks.

[0111] Fourth, this invention introduces a dynamic penalty weight coefficient during the node access sequence optimization process. This coefficient adaptively adjusts the impact of different constraints on the total delivery cost based on task urgency, material importance, delivery demand type, and time window margin. This allows delivery tasks with high urgency, high importance, or strict time requirements to obtain a more reasonable scheduling priority in multi-node access sequence optimization, thereby improving the matching degree between the distributed factory UAV multi-task delivery path planning results and actual production delivery needs.

[0112] Fifth, this invention employs an improved artificial lemming algorithm to optimize the target node access sequence and segmented delivery trajectory, and performs smoothing, constraint re-verification, and sequential splicing on the segmented delivery trajectory. This reduces the risk of the smoothed trajectory entering obstacle areas, no-fly zones, or violating UAV flight attitude constraints, thereby improving the executability of the target multi-node delivery path. Attached Figure Description

[0113] Figure 1 A flowchart illustrating the distributed factory area drone multi-node delivery path planning method provided in an embodiment of the present invention; Figure 2 This is a schematic diagram of a distributed two-dimensional distribution environment model for a factory area provided in an embodiment of the present invention; Figure 3 This is a schematic diagram of a distributed three-dimensional delivery environment model of a factory area provided in an embodiment of the present invention; Figure 4 This is a schematic diagram of the solution process of the improved artificial lemming algorithm provided in an embodiment of the present invention; Figure 5 This is a schematic diagram of the segmented delivery trajectory smoothing and constraint re-verification process provided in an embodiment of the present invention; Figure 6 This is a schematic diagram of the structure of a distributed factory area drone multi-node delivery route planning system provided in an embodiment of the present invention; Figure 7 This is a schematic diagram of the target node access sequence and segmented delivery trajectory planning results obtained based on the improved artificial lemming algorithm, as provided in an embodiment of the present invention. Detailed Implementation

[0114] The embodiments of the present invention will now be described with reference to the accompanying drawings. It should be understood that the following embodiments are merely illustrative of the technical solutions of the present invention and are not intended to limit the scope of protection of the present invention. Other embodiments obtained by those skilled in the art based on the present invention without creative effort should all fall within the scope of protection of the present invention.

[0115] In one embodiment, the present invention provides a distributed factory area UAV multi-node delivery path planning method. This method can be executed by a factory area logistics scheduling platform, a UAV ground control station, an edge computing device, or a server. It is used to determine the target node access sequence for a UAV to sequentially visit multiple delivery nodes in a distributed factory area, given multiple factory area functional nodes and multiple delivery tasks, and to generate segmented delivery tracks between adjacent delivery nodes.

[0116] like Figure 1 As shown, the process first acquires information on the distributed factory's delivery environment, drone operating parameters, and delivery tasks. The delivery environment information includes factory workspace boundaries, factory functional node information, obstacle area information, and no-fly zone information. Drone operating parameters include maximum payload, maximum range, permissible flight altitude range, maximum permissible yaw angle, maximum permissible pitch angle, and cruise speed. Delivery task information includes pickup nodes, delivery nodes, material payload, task priority, delivery time window, and node service demand type.

[0117] After obtaining the above information, a distributed 3D delivery space model of the factory area is constructed. This model includes the factory operation space, a set of factory functional nodes, a set of obstacle areas, a set of no-fly zones, and areas where drones can fly.

[0118] In one alternative implementation, the factory workspace is represented as follows:

[0119]

[0120] Where x, y, and z represent the spatial coordinates of the UAV in the three-dimensional coordinate system. , These represent the horizontal boundaries of the factory's delivery operation space. These represent the minimum and maximum altitudes at which drones are permitted to fly within the factory area, respectively.

[0121] The set of functional nodes in the factory area is represented as follows:

[0122]

[0123] in, Indicates the take-off and landing nodes of the drone. This represents the i-th functional node in the factory area. Each of the aforementioned factory area functional nodes includes at least node spatial coordinates, node type, and node service attributes; the node type includes at least one of the following: warehousing node, production workshop node, quality inspection node, assembly node, temporary parking node, and charging node;

[0124] The set of obstacle regions is represented as follows:

[0125]

[0126] in, This represents the obstacle area in the k-th factory area. The obstacle region is represented by a cuboid enclosure, a cylinder enclosure, a sphere enclosure, a polyhedron enclosure, or a combination thereof;

[0127] The set of no-fly zones is represented as follows:

[0128]

[0129] in, This represents the r-th no-fly zone. The no-fly zone includes at least one of the following: densely populated areas, hazardous materials storage areas, production control areas, and temporary control areas.

[0130] The area where the drone can fly is the area obtained by removing the set of obstacle areas and the set of no-fly zones from the factory's operating space, and is represented as follows:

[0131]

[0132] in, Indicates the area where drones are permitted to fly. This represents the set of three-dimensional factory work spaces defined by the spatial boundaries of the factory work space. Let m represent the k-th obstacle region, and m represent the number of obstacle regions. Let r represent the r-th no-fly zone, and q represent the number of no-fly zones; the flight area that the UAV can pass through is used to constrain the UAV's candidate waypoints, segmented delivery paths, and target multi-node delivery routes.

[0133] like Figure 2 and Figure 3 As shown, the above modeling method can map multiple functional nodes, obstacle areas, no-fly zones, and passable flight areas within the distributed factory area into a unified three-dimensional delivery space model, providing a spatial basis for subsequent node selection, cost calculation, and trajectory planning.

[0134]

[0135] in, This represents the s-th delivery task. And the task Represented as:

[0136]

[0137] in, This represents the pickup node for the s-th delivery task. This represents the delivery node corresponding to the s-th delivery task. This indicates the material load corresponding to the s-th delivery task. This indicates the task priority corresponding to the s-th delivery task. This represents the delivery time window corresponding to the s-th delivery task. This indicates the node service demand type corresponding to the s-th delivery task. The node service demand type includes at least one of the following: pickup, delivery, quality inspection and transfer, assembly and transfer, temporary parking, charging and replenishment, and transit buffer.

[0138] Based on the pickup and delivery nodes corresponding to each delivery task in the delivery task set R, an initial task node set is determined from the factory area functional node set V. Further, based on the UAV operating parameters, node service attributes, and delivery task constraints, an auxiliary node set is determined from the factory area functional node set V. The auxiliary node set includes at least one of charging nodes, temporary docking nodes, and transit buffer nodes. Specifically, charging nodes can be determined as auxiliary nodes based on range constraints, temporary docking nodes can be determined based on obstacle avoidance or task connection requirements, and transit buffer nodes can be determined based on the connection relationship between multiple delivery tasks or load adjustment requirements. Subsequently, based on the initial task node set, the auxiliary node set, and the UAV take-off and landing nodes, a set of delivery nodes to be visited is generated, and task constraint parameters are configured for each delivery node in the set of nodes to be visited. The task constraint parameters include at least one of node service type, node service time, load change, delivery time window, task priority, whether docking is allowed, and whether charging is allowed. Therefore, multiple delivery tasks, multiple functional nodes in the distributed factory area, and drone operation constraints are transformed into a set of delivery nodes to be accessed and task constraint parameters, providing a data foundation for the subsequent construction of the path cost matrix between nodes and the determination of the target node access sequence.

[0139] After obtaining the set of delivery nodes to be visited, the path cost between any two delivery nodes in the set is calculated based on the distributed 3D delivery space model of the factory area, and the path cost matrix is ​​constructed. In one optional implementation, let the number of nodes in the set of delivery nodes to be visited be NT, and the path cost matrix is ​​expressed as:

[0140] in, This represents the path cost matrix between nodes. Indicates the delivery node to be visited. To the delivery node to be accessed The path cost between nodes. For each element in matrix C. When i≠j, Indicates that the drone is generated by a node Fly to the node The corresponding comprehensive passage cost; when i=j, It can be set to 0, or set to an inaccessible flag according to actual planning needs. To ensure that the path cost between nodes reflects the actual delivery constraints within the distributed plant area, the path cost between nodes... It is not determined solely by the straight-line distance between two nodes, but rather by a comprehensive consideration of factors such as flight distance, flight undulations, spatial avoidance, time window, payload, and endurance. It is expressed as follows: in, Represents a node To node The estimated flight distance, Represents a node To node The cost of flight fluctuations, Represents a node To node The cost of space avoidance, Represents a node To node Time window penalty cost Represents a node To node The penalty cost of load constraint Represents a node To node The penalty cost of limited battery life; , , , , , These represent the weighting coefficients for the corresponding cost terms. Among them, the estimated flight distance... Based on nodes With nodes The spatial distance between the nodes, the length of the candidate detour path, or the estimated feasible track length are determined. This applies when there are no obstacle areas or no-fly zones between the two nodes. It can be determined based on the three-dimensional coordinate distance between the two nodes; when there is an obstacle area or a no-fly zone between the two nodes, The cost of flight fluctuations can be determined based on the length of the candidate path after detour. Used to reflect nodes To node The altitude changes between candidate paths can be determined based on the cumulative altitude difference between adjacent waypoints, thus reducing the additional energy consumption and attitude changes caused by frequent takeoffs and landings during UAV flight. Spatial avoidance costs. The time window penalty is determined based on the spatial relationship between the candidate flight path and the obstacle area set and the no-fly zone set. When a candidate flight path approaches an obstacle area or no-fly zone, the spatial avoidance cost is increased; when a candidate flight path crosses an obstacle area or no-fly zone, the connection between the corresponding nodes can be deemed infeasible, or a significant penalty cost can be imposed. Load constraint penalty cost and the penalty cost of range constraints The path cost matrix C is determined based on the delivery task's time requirements, the drone's load status during node access, and the drone's remaining flight time. Introducing these penalty terms avoids generating node access schemes that, while spatially accessible, fail to meet time window, load, or flight time requirements during task execution. Therefore, the inter-node path cost matrix C quantifies the spatial reachability, flight cost, and task constraints between different delivery nodes within the distributed factory area, providing a computational foundation for subsequent target node access sequence optimization. After constructing the inter-node path cost matrix C, the node access order is further optimized based on the inter-node path cost matrix C and task constraint parameters to determine the target node access sequence for the drone to sequentially access multiple delivery nodes in the current delivery task. In one optional implementation, the target node access sequence is represented as: in, This represents the optimized target node access sequence. Let π represent the delivery node to be visited by the drone at the k-th visit location, and M represent the length of the node visit sequence. For any candidate node visit sequence π, its basic path cost can be calculated based on the path cost between any two adjacent nodes in the sequence. Furthermore, to ensure the visit order meets the requirements of the actual factory delivery task, constraints such as pickup / delivery order, load, time window, and task priority are introduced on top of the basic path cost. The target node visit sequence is determined through the following optimization objective:

[0141] Where π represents the candidate node visit sequence. This represents the path cost between nodes in the candidate node access sequence, from the k-th access node to the (k+1)-th access node. This indicates a penalty for incorrect pickup / delivery order. This indicates the load constraint penalty term. This indicates the time window constraint penalty term. Indicates task priority penalty items; , , , These represent the weight coefficients of the corresponding penalty items.

[0142] Among them, the order of pickup and delivery penalty item This is used to restrict pickup nodes from being accessed before delivery nodes in the same delivery task. When a delivery node appears earlier than its corresponding pickup node in the candidate node access sequence, a pickup / delivery order penalty is added, or the candidate node access sequence is processed by node swapping, node insertion, or reordering. (Load constraint penalty) This is used to limit the payload of a drone during consecutive visits to multiple nodes, ensuring it does not exceed the maximum allowable payload. Specifically, the drone's current payload can be updated sequentially along the candidate node visit sequence: increasing the corresponding material payload when visiting a pickup node, and decreasing the corresponding material payload when visiting a delivery node; if the current payload at any visited location exceeds the maximum allowable payload, a payload constraint penalty is added. (Time window constraint penalty) This constraint ensures that the time it takes for a drone to arrive at or serve a corresponding delivery node meets the delivery task time requirements. The arrival time of the drone at each node can be estimated based on the path cost between nodes, the drone's cruise speed, and the node service time. A time window penalty is then determined based on the degree to which the arrival time deviates from the delivery time window. Task priority penalty is also included. This is used to reflect the difference in urgency between different delivery tasks. When the pickup or delivery node corresponding to a high-priority delivery task is too far down the candidate access sequence, or causes a significant delay in the task's completion time, a task priority penalty is added to guide the optimization process to prioritize high-priority delivery tasks.

[0143] The dynamic penalty weight coefficient can be determined based on the base weight and the task status adjustment factor:

[0144]

[0145] Where m represents any penalty item among the following: delivery order, load capacity, time window, or task priority. This represents the base weight of the m-th type of penalty term. This represents the adjustment coefficient for the m-th type of penalty term. This represents the task state adjustment factor corresponding to the candidate node access sequence π. The task state adjustment factor is determined by normalization based on at least one of the following: the urgency of the delivery task, the importance level of the material, the type of delivery demand, the remaining margin of the delivery time window, the current load status, and the degree of constraint violation.

[0146] Before evaluating the fitness of the candidate node access sequence, a dynamic penalty weight coefficient is generated based on the current delivery task status. Specifically, for each delivery task in the delivery task set, the task priority, material importance level, delivery demand type, and delivery time window are obtained. The remaining margin of the delivery time window is calculated based on the current time or the expected arrival time. The remaining margin of the delivery time window is used to characterize the available time remaining between the corresponding delivery task and its delivery time window deadline. The remaining margin of the delivery time window is determined based on the time difference between the delivery time window deadline and the current time, the expected arrival time, or the expected service completion time. The smaller the remaining margin of the delivery time window, the more pressing the time constraint of the corresponding delivery task.

[0147] When the remaining margin of the delivery time window is less than a preset threshold, the corresponding delivery task is determined to be in an emergency delivery state, and the dynamic penalty weight coefficient corresponding to the time window constraint penalty item is increased. When the task priority or material importance level of a delivery task exceeds a preset threshold, the dynamic penalty weight coefficient corresponding to the task priority penalty item is increased. When the material load corresponding to the delivery task is large, or when the candidate node access sequence makes the drone's current load close to the maximum allowable load, increase the dynamic penalty weight coefficient corresponding to the load constraint penalty term. When a delivery node in the candidate node access sequence arrives earlier than its corresponding pickup node, the dynamic penalty weight coefficient corresponding to the pickup / delivery order penalty term is increased. Alternatively, the candidate node access sequence may be determined as an infeasible sequence.

[0148] Each time the fitness of a candidate node access sequence is evaluated, the dynamic penalty weight coefficient is updated according to the task completion order, estimated arrival time, current load status and constraint violation status of the candidate node access sequence. The updated dynamic penalty weight coefficient is then substituted into the total delivery cost function to obtain the fitness function value of the candidate node access sequence.

[0149] Therefore, during the optimization process, the algorithm can adaptively adjust the penalty intensity of different constraint terms according to the changes in the delivery task status, so that delivery tasks with high urgency, high importance, or small remaining time window margin can obtain higher scheduling priority in the node access sequence optimization process, while avoiding the problem of urgent tasks being delayed or overloaded tasks being arranged unreasonably due to fixed penalty weights.

[0150] After determining the optimization objectives of the path cost matrix between nodes and the target node visit sequence, an improved artificial lemming algorithm is further used for optimization. For example... Figure 4 As shown, the improved artificial lemming algorithm can be used for a two-level solution process: first, to determine the target node access sequence in the set of delivery nodes to be visited; second, to perform segmented optimization of the three-dimensional flight path between two adjacent delivery nodes after the target node access sequence is determined. When used for node access sequence optimization, one candidate individual in the improved artificial lemming algorithm corresponds to one candidate node access sequence; when used for segmented path optimization, one candidate individual in the improved artificial lemming algorithm corresponds to a set of candidate path point sequences. Thus, the same optimization solution module can be adapted to both discrete node access sequence optimization problems and continuous three-dimensional path point optimization problems. In an optional implementation, an improved chaotic mapping method is first used to initialize the lemming population to generate initial candidate solutions. The improved chaotic mapping method can be an improved Tent chaotic mapping that introduces a random perturbation term, expressed as:

[0151]

[0152] Where k represents the number of mapping iterations, This represents the chaotic sequence value obtained by the k-th mapping, and rand(0,1) represents a random number between [0,1]. This represents the random disturbance intensity coefficient, and normalizes chaotic sequence values ​​that exceed the [0,1] interval.

[0153] Mapping the chaotic sequence values ​​generated by the improved Tent chaotic mapping to the search space yields the initial positions of individual lemmings:

[0154]

[0155] in, This represents the initial position of the i-th lemming individual. Let represent the chaotic sequence value used to generate the i-th lemming individual, and lb and ub represent the lower and upper bounds of the search space, respectively. When the improved artificial lemming algorithm is used for node access sequence optimization, the initial position of the lemming individual can be used as a random key vector, and candidate node access sequences can be generated based on the sorting results of the elements in the random key vector. For example, for a task containing multiple delivery nodes to be visited, a random key vector corresponding to the number of delivery nodes to be visited can be generated first, and then the corresponding nodes can be arranged in ascending order of random key values ​​to obtain a candidate node access sequence. For candidate node access sequences that do not meet the constraints of pickup / delivery order, load, time window, or endurance, their fitness can be evaluated through a penalty term, or the legality of the node sequence can be corrected. When the improved artificial lemming algorithm is used for segmented trajectory optimization, the initial position of the lemming individual can be mapped to a candidate trajectory point sequence. Each trajectory point in the candidate trajectory point sequence includes three-dimensional coordinate information and is subject to constraints such as the factory operation space, the drone's traversable flight area, obstacle areas, no-fly zones, and the drone's flight attitude. After population initialization, the current globally optimal individual is determined based on the fitness function value corresponding to the candidate solutions. Specifically, when the improved artificial lemming algorithm is used for node access sequence optimization, the fitness function can be determined based on the aforementioned total delivery cost; when the improved artificial lemming algorithm is used for segmented trajectory optimization, the fitness function can be determined based on the segmented trajectory length, flight fluctuation cost, and trajectory constraint penalty cost. Further, during the iterative search process, the lemming individual is controlled to perform either global exploration behavior or local development behavior according to the stage discrimination mechanism of the artificial lemming algorithm. Global exploration behavior is used to expand the coverage of candidate solutions in the search space to reduce the probability of the algorithm prematurely converging to a local optimum; local development behavior is used to refine the search near the current better solution to improve the optimization accuracy of the target node access sequence or segmented trajectory. After entering local development behavior, a physical strength judgment mechanism is introduced to adjust the local search strategy of the lemming individual. In one optional implementation, physical strength parameters are generated for the current lemming individual:

[0156]

[0157] in, Let represent the physical strength parameters of the i-th lemming individual. Represents a random number within the interval [0,1]. and These represent the physical fitness parameter adjustment coefficients;

[0158] Physical fitness parameters Compared with the preset physical strength threshold The comparison is performed, and the local search strategy for the current lemming individual is determined based on the comparison results;

[0159] Among them, when When, control the current lemming individual to execute a local foraging search strategy; when At that time, control the current lemming individual to execute a stamina-restoring search strategy;

[0160] Furthermore, based on the aforementioned physical strength parameters Determine local search weights:

[0161]

[0162] in, This represents the local search weight of the i-th lemming individual. This represents the weighting adjustment coefficient. Represents the optimal physical state parameters. Represents physical strength fluctuation parameters; local search weights Used to adjust the search step size and perturbation intensity in the local foraging search strategy or the physical recovery search strategy.

[0163] Through the aforementioned physical strength assessment mechanism, the search behavior can be adjusted based on the individual lemming's state during the local development phase, enabling the algorithm to have more flexible local optimization capabilities when approaching better node access sequences or better segmented tracks. After the lemming individual completes its position update, a cross-sectional optimization strategy is further introduced to perform secondary optimization on the updated candidate solutions. This cross-sectional optimization strategy includes lateral cross-operations and vertical cross-operations. The lateral cross-operation is used to reorganize information between corresponding dimensions of different lemming individuals. For the values ​​of the i-th and j-th lemming individuals in the d-th dimension, the lateral cross-sectional candidate solution is represented as:

[0164]

[0165]

[0166] in, and Let represent the values ​​of the i-th and j-th lemming individuals in the d-th dimension, respectively. and Let represent the values ​​of the candidate solutions generated after lateral crossover in the d-th dimension. and A random number within the interval [0,1]. and This is the cross disturbance coefficient;

[0167] The vertical crossover operation is used to perturb information between different dimensions of the same lemming individual, for two different dimensions in the i-th lemming individual. and The vertical cross candidate solution is represented as:

[0168]

[0169] in, This indicates that the candidate solutions generated after vertical crossover are in dimension The value on, It represents a random number within the interval [0,1].

[0170] Fitness evaluation is performed on the lateral and longitudinal crossover candidate solutions, and individuals are selected for retention from the pre-crossover lemming individuals, lateral crossover candidate solutions, and longitudinal crossover candidate solutions based on the fitness function value. Specifically, when the improved artificial lemming algorithm is used for node access order optimization, node sequence validity correction is performed on the lateral and longitudinal crossover candidate solutions; when the improved artificial lemming algorithm is used for segmented trajectory optimization, trajectory constraint validity correction is performed on the lateral and longitudinal crossover candidate solutions.

[0171] After completing the lateral and longitudinal crossover operations, the fitness of the candidate solutions generated by the crossover is evaluated, and individuals are selected for retention from the lemming individuals before crossover, the lateral crossover candidate solutions, and the longitudinal crossover candidate solutions based on the fitness function value. Further, the validity of the retained individuals is corrected according to the different optimization objects. When used for node access sequence optimization, candidate solutions are deduplicated, missing nodes are filled in, delivery order is corrected, auxiliary nodes are inserted, or constraint penalties are applied to obtain a valid candidate node access sequence. When used for segmented trajectory optimization, candidate solutions are corrected for search boundary, altitude constraint, obstacle and no-fly zone avoidance, yaw angle, and pitch angle to obtain candidate segmented trajectories that satisfy the factory area space constraints and UAV flight constraints. When a preset stopping condition is met, the iterative search stops, and the current globally optimal individual is output. The preset stopping conditions include reaching the maximum number of iterations, the fitness change being less than a preset threshold for multiple consecutive iterations, or the current optimal solution meeting a preset delivery cost requirement. The current globally optimal individual serves as the target node access sequence during the node access sequence optimization process, and as the segmented delivery track between two adjacent delivery nodes to be accessed during the segmented track optimization process.

[0172] After determining the target node access sequence, the three-dimensional flight path between two adjacent delivery nodes to be accessed is segmented and planned according to the target node access sequence.

[0173] Specifically, if the target node access sequence is Then, the two adjacent delivery nodes to be visited will be... and As the starting and ending points of a segmented trajectory planning task, solutions are obtained separately. to , to until to The three-dimensional flight path between nodes is thus transformed into a segmented flight path planning problem between multiple adjacent nodes.

[0174] For each segmented flight path planning task, the two corresponding delivery nodes to be visited are used as the segment start and end points. A sequence of candidate waypoints is generated within the UAV's navigable flight area, and an improved artificial lemming algorithm is used to optimize the candidate waypoint sequence. The candidate waypoint sequence should meet constraints such as factory operation space boundaries, obstacle avoidance, no-fly zone avoidance, cruising altitude, maximum permissible yaw angle, and maximum permissible pitch angle.

[0175] In one optional implementation, the fitness function of the segmented trajectory is determined based on the segmented trajectory length, flight fluctuation cost, and trajectory constraint penalty cost. The segmented trajectory length characterizes the flight distance between two adjacent delivery nodes to be visited; the flight fluctuation cost characterizes the degree of trajectory altitude variation; and the trajectory constraint penalty cost characterizes whether the candidate trajectory enters an obstacle area, a no-fly zone, or violates UAV flight attitude constraints.

[0176] Through the optimization process described above, multiple segmented delivery tracks can be obtained. Each segmented delivery track includes a segment start point, a segment end point, and a sequence of trackpoints located between the segment start point and the segment end point.

[0177] Furthermore, such as Figure 5 As shown, multiple segmented delivery tracks are smoothed, constraint re-verified, and sequentially spliced.

[0178] Since the segmented delivery tracks output by the improved artificial lemming algorithm are typically composed of discrete waypoints, directly connecting these discrete waypoints may lead to abrupt changes in local track transitions, excessive changes in flight attitude, or insufficient track continuity. Therefore, after obtaining the segmented delivery tracks, each segment is smoothed to generate a smoothed, continuous segmented track.

[0179] In one alternative implementation, curve smoothing can be achieved using B-spline curves, Bézier curves, polynomial curves, or combinations thereof. During curve smoothing, discrete track points in the segmented delivery track can be used as control points or fitting points to generate a continuous curve connecting the start and end points of the segments.

[0180] After curve smoothing, the smoothed continuous segmented tracks are discretely sampled, and the track segments formed by the sampled points and adjacent sampled points are subjected to constraint re-verification. The constraint re-verification includes at least one of the following: passable flight area verification, obstacle area verification, no-fly zone verification, cruise altitude verification, maximum permissible yaw angle verification, and maximum permissible pitch angle verification.

[0181] Among them, the passable flight area verification is used to determine whether the sampling point is located within the passable flight area of ​​the UAV; the obstacle area verification and no-fly zone verification are used to determine whether the track segment formed by the sampling point or adjacent sampling points has spatial interference with the obstacle area or no-fly zone; the cruise altitude verification is used to determine whether the altitude of the sampling point is within the allowable flight altitude range; and the maximum allowable yaw angle verification and the maximum allowable pitch angle verification are used to determine whether the flight attitude change between adjacent track segments meets the maneuverability requirements of the UAV.

[0182] If the smoothed continuous segmented track satisfies the above constraint re-verification, then the continuous segmented track is considered a valid segmented track. If the smoothed continuous segmented track does not satisfy the constraint re-verification, then the local track segments that do not satisfy the constraints are identified, and auxiliary track points are inserted into these local track segments. Subsequently, based on the track point sequence after inserting auxiliary track points, curve smoothing and constraint re-verification are performed again until a continuous segmented track that satisfies the constraint requirements is obtained, or a preset correction stopping condition is reached.

[0183] After all consecutive segmented tracks have satisfied the constraint re-verification, the multiple consecutive segmented tracks are sequentially concatenated according to the target node access sequence. During concatenation, the endpoint of the previous consecutive segmented track should correspond to the same delivery node to be accessed as the starting point of the next consecutive segmented track. After concatenation is completed, the target multi-node delivery path is generated.

[0184] The target multi-node delivery path includes the target node access sequence, segmented trajectory information between adjacent delivery nodes to be accessed, smoothed continuous trajectory information, and constraint re-verification results. This target multi-node delivery path can be output to the UAV flight control system, the factory logistics scheduling platform, or a visualization display terminal for executing or displaying UAV multi-node delivery tasks within a distributed factory area.

[0185] like Figure 6 As shown, this embodiment also provides an electronic device, which includes a processor, a memory, and a computer program stored in the memory and executable by the processor. When the processor executes the computer program, it implements the steps of the above-described method for multi-node delivery path planning of unmanned aerial vehicles (UAVs) in a distributed factory area.

[0186] The memory is used to store information on the factory area's operational space, functional nodes, obstacle areas, no-fly zones, delivery tasks, UAV flight constraint parameters, path cost matrices between nodes, and path planning algorithms. The processor is used to perform operations such as node information acquisition, construction of passable flight areas, generation of delivery task node sets, calculation of path costs between nodes, determination of target node access sequences, segmented trajectory optimization, constraint verification, and output of target multi-node delivery paths.

[0187] The electronic device can be a server, an industrial computer, an edge computing device, or a UAV ground control terminal. Figure 6 The structures shown are merely examples and should not be construed as limiting the specific hardware structure of the electronic device of the present invention.

[0188] In one alternative implementation, such as Figure 7 As shown, the target multi-node delivery path can be output visually. The visualization output includes the drone's takeoff and landing nodes, delivery nodes to be visited, the target node visit sequence, segmented delivery trajectories, and the factory's spatial constraints. Therefore, factory logistics dispatchers can confirm the drone's visit order and flight path among multiple delivery nodes based on the visualization output.

[0189] In one alternative implementation, such as Figure 7 As shown, after optimizing the target node access sequence and planning the segmented delivery trajectory, the system outputs the multi-node delivery path planning results based on the improved artificial lemming algorithm. The results demonstrate that the UAV starts from the take-off and landing node, sequentially visits multiple delivery nodes according to the target node access sequence, and generates segmented delivery trajectories between adjacent delivery nodes that satisfy obstacle avoidance, no-fly zone avoidance, and flight attitude constraints. Thus, the node access sequence optimization results and the 3D trajectory planning results can be presented in a visual manner.

[0190] The system includes an environment modeling module, a task node generation module, a node cost matrix construction module, a node access sequence optimization module, an improved artificial lemming algorithm solution module, a segmented trajectory planning module, a path processing module, and a path output module.

[0191] The environmental modeling module is used to acquire information on the distribution environment of the distributed factory area, drone operating parameters, and delivery task information, and to construct a three-dimensional distribution space model of the distributed factory area. This three-dimensional distribution space model includes the factory area's operational space, a set of functional nodes, a set of obstacle areas, a set of no-fly zones, and areas where drones can fly.

[0192] The task node generation module is connected to the environment modeling module. It determines an initial task node set from the set of functional nodes in the factory area based on delivery task information, and determines an auxiliary node set based on UAV operating parameters, node service attributes, and delivery task constraints. The task node generation module further generates a set of delivery nodes to be accessed based on the initial task node set, the auxiliary node set, and the UAV take-off and landing nodes, and configures task constraint parameters for each delivery node to be accessed.

[0193] The inter-node cost matrix construction module is connected to the task node generation module. It is used to calculate the inter-node path cost between any two delivery nodes in the set of delivery nodes to be visited, and to construct the inter-node path cost matrix. The inter-node path cost is determined based on the estimated flight distance between nodes, flight fluctuation cost, spatial avoidance cost, time window penalty cost, load constraint penalty cost, and endurance constraint penalty cost.

[0194] The node access sequence optimization module is connected to the inter-node cost matrix construction module. It optimizes the node access order of the set of delivery nodes to be accessed based on the inter-node path cost matrix and task constraint parameters, determining the target node access sequence. The target node access sequence represents the order in which the drone sequentially accesses multiple delivery nodes to be accessed.

[0195] The improved artificial lemming algorithm solution module is connected to both the node visit sequence optimization module and the segmented trajectory planning module, and is used to perform optimization solutions based on the improved artificial lemming algorithm. During the node visit sequence optimization process, the improved artificial lemming algorithm solution module is used to generate and update candidate node visit sequences; during the segmented trajectory planning process, the improved artificial lemming algorithm solution module is used to generate and update candidate waypoint sequences.

[0196] The segmented trajectory planning module is connected to the node access sequence optimization module. It is used to determine two adjacent delivery nodes to be visited as a segmented trajectory planning task according to the target node access sequence, and to perform three-dimensional flight trajectory planning for each segmented trajectory planning task to obtain multiple segmented delivery trajectories.

[0197] The path processing module is connected to the segmented trajectory planning module and is used to perform curve smoothing, discrete sampling, constraint re-verification, and sequential splicing on multiple segmented delivery trajectories to generate a target multi-node delivery path. The constraint re-verification includes at least one of the following: passable flight area verification, obstacle area verification, no-fly zone verification, cruising altitude verification, maximum permissible deflection angle verification, and maximum permissible pitch angle verification.

[0198] The path output module is connected to the path processing module and is used to output the target multi-node delivery path. The target multi-node delivery path includes the target node access sequence, segmented track information between adjacent delivery nodes to be accessed, smoothed continuous track information, and constraint re-verification results.

[0199] In this embodiment, the distributed 3D delivery space model of the factory area output by the environmental modeling module serves as the spatial basis for task node generation, inter-node cost calculation, and segmented trajectory planning; the set of delivery nodes to be visited and task constraint parameters output by the task node generation module serve as the data basis for constructing the inter-node cost matrix and optimizing the node access order; the target node access sequence output by the node access sequence optimization module serves as the input of the segmented trajectory planning module; and the path processing module performs post-processing on the multiple segmented delivery trajectories output by the segmented trajectory planning module to obtain a target multi-node delivery path that can be used for UAV delivery execution.

[0200] In another embodiment, the present invention also provides an electronic device. The electronic device may be a server, an edge computing device, a drone ground control device, a factory logistics scheduling terminal, or other computing devices with data processing capabilities.

[0201] The electronic device includes a processor, a memory, and a computer program stored in the memory and capable of running on the processor. When the processor executes the computer program, it implements the aforementioned multi-node delivery path planning method for distributed factory areas.

[0202] Specifically, when the processor executes the computer program, it can acquire information on the distribution environment of the distributed factory area, the operating parameters of the UAV, and the delivery task information, and construct a three-dimensional distribution space model of the distributed factory area; determine the set of delivery nodes to be visited based on the delivery task information, and configure task constraint parameters for the delivery nodes to be visited; calculate the path cost between the delivery nodes to be visited and construct the path cost matrix between the nodes; determine the target node access sequence based on the path cost matrix between the nodes and the task constraint parameters; plan the segmented delivery trajectory between adjacent delivery nodes to be visited according to the target node access sequence; perform smoothing, constraint re-verification, and sequential splicing on multiple segmented delivery trajectories, and output the target multi-node delivery path.

[0203] The memory can be used to store data such as a distributed 3D delivery space model of the factory area, information on functional nodes of the factory area, information on obstacle areas, information on no-fly zones, UAV operating parameters, delivery task information, path cost matrix between nodes, target node access sequence, segmented delivery trajectory, and target multi-node delivery path.

[0204] The electronic device may also include a communication interface. This communication interface is used to interact with the factory logistics scheduling platform, the UAV flight control system, the UAV ground control station, sensing devices, or a visual display terminal. Through this communication interface, the electronic device can receive delivery task information and factory environment information, and output the target multi-node delivery path to the UAV flight control system or the factory logistics scheduling platform.

[0205] In another embodiment, the present invention also provides a computer-readable storage medium. The computer-readable storage medium stores a computer program, which, when executed by a processor, implements the aforementioned method for multi-node delivery path planning for distributed factory areas using unmanned aerial vehicles (UAVs).

[0206] The computer-readable storage medium may include a read-only memory, random access memory, disk, optical disk, flash memory, mobile storage device, or other medium capable of storing a computer program. The computer program may include program instructions for performing environment modeling, task node generation, inter-node cost matrix construction, node access order optimization, improved artificial lemming algorithm solution, segmented trajectory planning, path processing, and path output.

[0207] It should be noted that the technical features in the aforementioned method implementations, the module functions in the system implementations, and the program execution process in the electronic device and computer-readable storage medium implementations can be referred to each other. Any equivalent substitutions, combinations, or modifications made under the inventive concept to the order of steps, module division, data structure, algorithm parameters, or constraint handling methods should fall within the protection scope of this invention.

Claims

1. A distributed factory area drone multi-node delivery route planning method, characterized in that, Includes the following steps: The system acquires information on the distribution environment of the distributed factory area, drone operating parameters, and delivery task information to construct a three-dimensional distribution space model of the distributed factory area. The three-dimensional distribution space model of the distributed factory area includes multiple functional nodes of the factory area, obstacle areas, no-fly zones, and passable flight areas. Based on the delivery task information, a set of delivery nodes to be accessed is determined from the multiple factory area functional nodes, and task constraint parameters are configured for each delivery node to be accessed in the set of delivery nodes to be accessed. Based on the distributed three-dimensional delivery space model of the factory area, the path cost between any two delivery nodes to be accessed in the set of delivery nodes to be accessed is calculated, and the path cost matrix between nodes is constructed according to the path cost between nodes. Based on the path cost matrix between nodes and the task constraint parameters, an optimization solution method based on the improved artificial lemming algorithm is used to optimize the node access order of the set of delivery nodes to be visited, and determine the target node access sequence. According to the target node access sequence, the optimization solution method based on the improved artificial lemming algorithm is used to optimize the three-dimensional flight path between two adjacent delivery nodes to be visited, and obtain multiple segmented delivery paths. The multiple segmented delivery paths are smoothed, constraint re-verified, and sequentially spliced ​​to generate a target multi-node delivery path that satisfies the target node access sequence and the UAV flight constraints. Output the target multi-node delivery path, wherein the target multi-node delivery path includes the target node access sequence of the UAV and the segmented track information between adjacent delivery nodes to be accessed.

2. The distributed factory area UAV multi-node delivery path planning method according to claim 1, characterized in that, The distributed three-dimensional delivery space model of the factory area includes the factory operation space, the set of factory functional nodes, the set of obstacle areas, the set of no-fly zones, and the area where drones can pass and fly. The factory workspace is referred to as: Where x, y, and z represent the spatial coordinates of the UAV in the three-dimensional coordinate system. , These represent the horizontal boundaries of the factory's delivery operation space. These represent the minimum and maximum altitudes at which drones are permitted to fly within the factory area, respectively. The set of functional nodes in the factory area is represented as follows: in, Indicates the take-off and landing nodes of the drone. This represents the i-th functional node in the factory area. Each of the aforementioned factory area functional nodes includes at least node spatial coordinates, node type, and node service attributes; the node type includes at least one of the following: warehousing node, production workshop node, quality inspection node, assembly node, temporary parking node, and charging node; The set of obstacle regions is represented as follows: in, This represents the obstacle area in the k-th factory area. The obstacle region is represented by a cuboid enclosure, a cylinder enclosure, a sphere enclosure, a polyhedron enclosure, or a combination thereof; The set of no-fly zones is represented as follows: in, This represents the r-th no-fly zone. The no-fly zone includes at least one of the following: densely populated areas, hazardous materials storage areas, production control areas, and temporary control areas. The area where the drone can fly is the area obtained by removing the set of obstacle areas and the set of no-fly zones from the factory's operating space, and is represented as follows: in, Indicates the area where drones are permitted to fly. This represents the set of three-dimensional factory work spaces defined by the spatial boundaries of the factory work space. Let m represent the k-th obstacle region, and m represent the number of obstacle regions. Let r represent the r-th no-fly zone, and q represent the number of no-fly zones; the flight area that the UAV can pass through is used to constrain the UAV's candidate waypoints, segmented delivery paths, and target multi-node delivery routes.

3. The distributed factory area UAV multi-node delivery path planning method according to claim 1, characterized in that, The delivery task information includes a set of delivery tasks: in, This represents the s-th delivery task. And the delivery task Represented as: in, This represents the pickup node for the s-th delivery task. This represents the delivery node corresponding to the s-th delivery task. This indicates the material load corresponding to the s-th delivery task. This indicates the task priority corresponding to the s-th delivery task, which is determined based on at least one of the following: the urgency of the delivery task, the importance level of the materials, the type of delivery demand, and the remaining margin of the delivery time window. This represents the delivery time window corresponding to the s-th delivery task. This indicates the node service demand type corresponding to the s-th delivery task, and the node service demand type includes at least one of pickup, delivery, quality inspection flow, assembly flow, temporary parking, charging and replenishment, and transit buffer. Based on the pickup and delivery nodes corresponding to each delivery task in the delivery task set R, an initial task node set is determined from the factory area functional node set, and an auxiliary node set is determined from the factory area functional node set based on the UAV operating parameters, node service attributes, and delivery task constraints. The set of delivery nodes to be visited is generated based on the initial task node set, the auxiliary node set, and the drone take-off and landing nodes; wherein, the auxiliary node set includes at least one of charging nodes, temporary docking nodes, and transit buffer nodes, and the nodes in the auxiliary node set are determined as auxiliary nodes to participate in the current path planning based on endurance constraints, obstacle avoidance requirements, or task connection requirements.

4. The distributed factory area UAV multi-node delivery path planning method according to claim 1, characterized in that, The path cost matrix between nodes is represented as follows: in, This represents the path cost matrix between nodes. This represents the nodes in the set of delivery nodes to be accessed. To node The path cost between nodes, This indicates the number of nodes in the set of delivery nodes to be accessed; The path cost between nodes The cost is determined by a weighted average of the estimated flight distance between nodes, flight fluctuation cost, spatial avoidance cost, time window penalty cost, load constraint penalty cost, and endurance constraint penalty cost, and is expressed as follows: in, Represents a node To node The estimated flight distance, Represents a node To node The cost of flight fluctuations, Represents a node To node The cost of space avoidance, Represents a node To node Time window penalty cost Represents a node To node The penalty cost of load constraint Represents a node To node The penalty cost of limited battery life , , , , These are the weighting coefficients for the corresponding cost items; The cost of space avoidance According to the node To node The safe distance relationship between the candidate flight path and the set of obstacle areas and the set of no-fly zones is determined; the time window penalty cost Load constraint penalty cost and the penalty cost of range constraints The determination is based on the delivery time window, the drone's payload status, and the drone's remaining battery life, respectively.

5. The distributed factory area UAV multi-node delivery path planning method according to claim 1, characterized in that, The target node access sequence is represented as follows: in, This represents the optimized target node access sequence. This represents the delivery node to be visited corresponding to the kth visit location of the drone, and M represents the length of the node visit sequence. The target node access sequence is determined by minimizing the total delivery cost, which is determined based on the path cost between nodes, pickup and delivery order constraints, load constraints, time window constraints, and task priority constraints. Its optimization objective is expressed as: in, This represents the candidate node access sequence. This represents the path cost between two adjacent nodes in the candidate node visit sequence. This indicates a penalty for incorrect pickup / delivery order. This indicates the load constraint penalty term. This indicates the time window constraint penalty term. This indicates a task priority penalty item. , , , These represent the dynamic penalty weight coefficients for the corresponding penalty items; the dynamic penalty weight coefficients are determined based on at least one of the following: the urgency of the delivery task, the importance level of the materials, the type of delivery demand, the remaining margin of the delivery time window, the current load status of the drone, and the degree of constraint violation of the candidate node access sequence. The pickup and delivery order constraint is used to restrict the pickup node from being accessed before the delivery node in the same delivery task. The load constraint is used to restrict the load of the drone during the node access process from not exceeding the maximum allowable load. The time window constraint is used to restrict the time of the drone to arrive at or serve the corresponding node to meet the time requirements of the delivery task. The task priority constraint is used to enable the node corresponding to the higher priority delivery task to obtain a higher access priority in the node access order optimization.

6. The distributed factory area UAV multi-node delivery path planning method according to claim 5, characterized in that, The optimization solution method based on the improved artificial lemming algorithm is used to optimize the solution of the target node access sequence and the three-dimensional flight path between the two adjacent delivery nodes to be visited; The improved artificial lemming algorithm includes: An improved chaotic mapping method is used to initialize the lemming population to generate candidate node access sequences for node access order optimization and candidate waypoint sequences for segmented path optimization. The current global optimal individual is determined based on the fitness function value corresponding to the candidate solution. When the improved artificial lemming algorithm is used for node access sequence optimization, the fitness function is determined based on the total delivery cost with introduced dynamic penalty weight coefficients. When the improved artificial lemming algorithm is used for segmented trajectory optimization, the fitness function is determined based on the segmented trajectory length, flight fluctuation cost, and trajectory constraint penalty cost. During the iterative search process, the lemming individual is controlled to perform global exploration behavior or local development behavior according to the stage discrimination mechanism of the artificial lemming algorithm, and a physical strength judgment mechanism is introduced in the local development behavior to adjust the local search strategy according to the physical strength of the lemming individual. After the lemming individuals complete their position updates, a cross-sectional optimization strategy is introduced to perform secondary optimization on the updated candidate solutions, and the candidate solutions before and after optimization are selected based on the fitness function value. When the preset stopping condition is met, the current global best individual is output; wherein, in the node access sequence optimization process, the candidate node access sequence corresponding to the current global best individual is used as the target node access sequence, and in the segmented trajectory optimization process, the candidate trajectory point sequence corresponding to the current global best individual is used as the optimization result of the segmented delivery trajectory.

7. The distributed factory area UAV multi-borrowing point delivery route planning method according to claim 6, characterized in that, The improved chaotic mapping method is an improved Tent chaotic mapping that introduces a random perturbation term. The improved Tent chaotic mapping is expressed as follows: Where k represents the number of mapping iterations, The chaotic sequence value obtained by the k-th mapping, rand(0,1) represents a random number between [0,1]. This represents the random disturbance intensity coefficient, and normalizes chaotic sequence values ​​that exceed the [0,1] interval. Mapping the chaotic sequence values ​​generated by the improved Tent chaotic mapping to the search space yields the initial positions of individual lemmings: in, This represents the initial position of the i-th lemming individual. Let represent the chaotic sequence value used to generate the i-th lemming individual, and lb and ub represent the lower bound and upper bound of the search space, respectively; When the improved artificial lemming algorithm is used for node access order optimization, the initial position of the lemming individual is used as a random key vector, and a candidate node access sequence is generated based on the sorting result of the random key vector; when the improved artificial lemming algorithm is used for segmented trajectory optimization, the initial position of the lemming individual is mapped to a candidate trajectory point sequence. The physical strength assessment mechanism includes: After the improved artificial lemming algorithm enters local development behavior, physical parameters are generated for the current lemming individual: in, Let represent the physical strength parameters of the i-th lemming individual. Represents a random number within the interval [0,1]. and These represent the adjustment coefficients for physical parameters; The physical parameters Compared with the preset physical strength threshold The comparison is performed, and the local search strategy for the current lemming individual is determined based on the comparison results; Among them, when When, control the current lemming individual to execute a local foraging search strategy; when At that time, control the current lemming individual to execute a stamina-restoring search strategy; According to the physical strength parameters Determine the local search weights: in, This represents the local search weight of the i-th lemming individual. This represents the weighting adjustment coefficient. Represents the optimal physical state parameters. Indicates parameters related to physical strength fluctuations; Based on the local search weights Adjust the search step size and perturbation intensity in the local foraging search strategy or the physical recovery search strategy; The cross-sectional optimization strategy includes lateral cross-sectional operations and vertical cross-sectional operations; The lateral crossover operation is used to reorganize information between corresponding dimensions of different lemming individuals. For the values ​​of the i-th and j-th lemming individuals in the d-th dimension, the lateral crossover candidate solution is represented as: in, and Let represent the values ​​of the i-th and j-th lemming individuals in the d-th dimension, respectively. and Let represent the values ​​of the candidate solutions generated after lateral crossover in the d-th dimension. and A random number within the interval [0,1]. and This is the cross disturbance coefficient; The vertical crossover operation is used to perturb information between different dimensions of the same lemming individual, for two different dimensions in the i-th lemming individual. and The vertical cross candidate solution is represented as: Fitness evaluation is performed on the lateral and longitudinal crossover candidate solutions, and individuals are selected for retention from the pre-crossover lemming individuals, lateral crossover candidate solutions, and longitudinal crossover candidate solutions based on the fitness function value. Specifically, when the improved artificial lemming algorithm is used for node access order optimization, node sequence validity correction is performed on the lateral and longitudinal crossover candidate solutions; when the improved artificial lemming algorithm is used for segmented trajectory optimization, trajectory constraint validity correction is performed on the lateral and longitudinal crossover candidate solutions.

8. The distributed factory area UAV multi-node delivery path planning method according to claim 1, characterized in that, The process of smoothing, constraining re-verification, and sequentially splicing multiple segmented delivery tracks includes: Based on the target node access sequence, the segmented delivery track between two adjacent delivery nodes to be accessed is taken as the track segment to be smoothed. The track segment to be smoothed is subjected to curve smoothing processing to obtain a smoothed continuous segmented track; Discrete sampling is performed on the smoothed continuous segmented trajectory, and constraint re-verification is performed on the trajectory segments formed by the sampling points and adjacent sampling points based on the UAV's passable flight area, obstacle area, no-fly zone and UAV flight attitude constraints. If the smoothed continuous segmented track does not meet the constraint re-verification, then auxiliary track points are inserted into the local track segments that do not meet the constraints, and the curve smoothing and constraint re-verification are performed again based on the track point sequence after the insertion of auxiliary track points. The continuous segmented tracks that satisfy the constraint re-verification are sequentially spliced ​​together according to the target node access sequence to generate the target multi-node delivery path. The UAV flight attitude constraints include at least one of the following: cruise altitude constraint, maximum allowable yaw angle constraint, and maximum allowable pitch angle constraint; the curve smoothing process is implemented using B-spline curves, Bézier curves, polynomial curves, or combinations thereof.

9. A distributed factory area drone multi-node delivery route planning system, characterized in that, The system is used to execute the UAV multi-node delivery path planning method for distributed factory areas as described in any one of claims 1 to 8, the system comprising: The environmental modeling module is used to acquire information on the distribution environment of the distributed factory area, the operating parameters of the drones, and the delivery task information, and to construct a three-dimensional delivery space model of the distributed factory area. The task node generation module is used to determine the set of delivery nodes to be accessed from multiple functional nodes in the distributed three-dimensional delivery space model of the factory area based on the delivery task information, and to configure task constraint parameters for each delivery node to be accessed in the set of delivery nodes to be accessed. The inter-node cost matrix construction module is used to calculate the inter-node path cost between any two delivery nodes in the set of delivery nodes to be visited, and to construct the inter-node path cost matrix based on the inter-node path cost. The node access sequence optimization module is used to optimize the node access order of the set of delivery nodes to be accessed based on the path cost matrix between nodes and the task constraint parameters, and to determine the target node access sequence for the UAV to access multiple delivery nodes in sequence. The node access sequence optimization module is also used to determine the dynamic penalty weight coefficient based on the urgency of the delivery task, the importance level of the materials, the type of delivery demand, the remaining margin of the delivery time window, the current load status of the UAV and the degree of constraint violation of the candidate node access sequence, and to determine the target node access sequence based on the dynamic penalty weight coefficient. An improved artificial lemming algorithm solution module is used to optimize the solution of the target node access sequence and the three-dimensional flight path between two adjacent delivery nodes to be visited based on the improved artificial lemming algorithm; The segmented trajectory planning module is used to plan the three-dimensional flight trajectory between two adjacent delivery nodes to be visited according to the target node access sequence, so as to obtain multiple segmented delivery trajectories. The path processing module is used to smooth, re-verify, and sequentially splice the multiple segmented delivery tracks to generate a target multi-node delivery path. The path output module is used to output the target multi-node delivery path.

Images