CN113706081A

CN113706081A - Unmanned aerial vehicle goods taking and delivering system and method based on urban roof automatic express device

Info

Publication number: CN113706081A
Application number: CN202111010035.XA
Authority: CN
Inventors: 伍国华; 洪芳宇
Original assignee: Central South University
Current assignee: Central South University
Priority date: 2021-08-31
Filing date: 2021-08-31
Publication date: 2021-11-26
Anticipated expiration: 2041-08-31
Also published as: CN113706081B

Abstract

The invention discloses an unmanned aerial vehicle goods taking and delivering system and method based on an automatic express delivery device on a city roof, wherein the method comprises the following steps: acquiring position information and goods taking and delivering information of multiple warehouses, multiple unmanned aerial vehicles and multiple customer points; designing a comprehensive scheduling problem of multiple centers, multiple customer points and multiple unmanned aerial vehicles; allocating an initial task to the single center, and reallocating the task according to operators in various fields; finding a path planning scheme for completing the task for each warehouse according to a plurality of neighborhood operators; continuously iterating and interacting the steps, and searching a global optimal solution until an algorithm termination condition is met; and distributing multiple unmanned aerial vehicles according to the global optimal solution to take and deliver goods from multiple warehouses and multiple customer points. The invention fully utilizes the vacant space of the city; the unmanned aerial vehicle can fly linearly on the way, and the taking and delivering efficiency is high; competition of space backlog and queuing with traditional goods taking and delivering is avoided; compared with the existing algorithm, the invention can generate a high-quality scheduling scheme in a reasonable time.

Description

Unmanned aerial vehicle goods taking and delivering system and method based on urban roof automatic express device

Technical Field

The invention belongs to the technical field of logistics, and particularly relates to an unmanned aerial vehicle goods taking and delivering system and method based on an automatic urban roof express delivery device.

Background

At present, most of researches on the application of unmanned aerial vehicles to package delivery problems focus on modes of independently using unmanned aerial vehicles or trucks and unmanned aerial vehicles, most of technical schemes focus on solving the package delivery problems, but with the increase of online shopping, reverse logistics caused by services such as goods returning and goods changing lack of scientific solutions, and few of technical schemes consider the package delivery and the package pickup problem; in more technical schemes, the mode of truck and unmanned aerial vehicle is adopted for solving the problem that packages in the last kilometer are obtained by the unmanned aerial vehicle, the mode of automatic access equipment and unmanned aerial vehicle is not adopted, and the mode of integral solution is adopted for solving the problem of package taking and delivering by the unmanned aerial vehicle. In the aspect of solving the large-scale task scheduling problem, solutions generated by a classical heuristic algorithm and a meta-heuristic algorithm are unsatisfactory, and a satisfactory solution is difficult to generate in a reasonable time.

Disclosure of Invention

In view of the above, the invention researches a brand-new mode, and applies an Unmanned Aerial Vehicle (UAV) to a logistics scene of the last kilometer considering both picking and delivering, and the scheme sets a picking and delivering point on the roof, so that customers can pick and send articles as conveniently as possible, and the automation degree of the last kilometer is improved. The perfection of the automatic storage equipment and the maturity of the unmanned aerial vehicle technology provide technical support for the invention. In this mode, a plurality of warehouses capable of automatically accessing packages are arranged on the roof, customer points are distributed under the warehouses according to the Euclidean distance, and the unmanned aerial vehicle cluster is used for simultaneously completing the fetching and delivering of the packages, so that the warehouse is a multi-row parallel scheduling and routing warehouse project (mD-PSTSP). The scheduling problem of the parcel picking and delivering task of multiple centers, multiple customer points and multiple unmanned aerial vehicles can be considered in consideration of multiple constraints such as range and load.

The invention constructs a mixed integer linear programming model of multi-warehouse, multi-unmanned aerial vehicle and multi-customer-point parcel pickup and delivery problems, provides a two-stage optimization framework to help solve the original problems, and finally designs an SATO-SVND algorithm.

The invention discloses an unmanned aerial vehicle cargo taking and delivering system based on an automatic express device on a city roof, which is characterized in that the automatic express device is arranged on the city roof, and the urban vacant space is utilized, so that the unmanned aerial vehicle can fly linearly, and the competition with the space backlog and queue of the traditional cargo taking and delivering system is avoided, and the system comprises:

the data input module is used for acquiring position information and goods taking and delivering information of a plurality of warehouses, a plurality of unmanned aerial vehicles and a plurality of customer points, and the customer points are automatic goods express devices arranged on a roof;

the scheduling module is used for designing a comprehensive scheduling problem based on unmanned aerial vehicle parcel taking and delivering, multi-center unmanned aerial vehicle load and range, multi-customer point and multi-unmanned aerial vehicle;

the task allocation module is used for allocating initial tasks to the single center and reallocating the tasks according to operators in various fields;

the path planning module finds a path planning scheme for completing the task for each warehouse according to the multiple neighborhood operators;

the optimization module continuously iterates the scheduling task allocation module and the path planning module, and searches for a global optimal solution until an algorithm termination condition is met; setting a small loop and a large loop, wherein the small loop generates a new scheduling scheme meeting constraint conditions, the large loop carries out local search on the basis of the scheduling scheme generated by the small loop to form a new scheduling scheme, and then determines whether to accept a result of the local search according to a greedy principle;

and the goods taking and delivering module is used for distributing multiple unmanned aerial vehicles according to the global optimal solution to take and deliver goods from multiple warehouses and multiple customer points.

Further, the unmanned aerial vehicle goes to the next customer point or warehouse after loading and unloading goods, or automatically changes the battery on the roof; for the customer points which only take goods, arranging an unmanned aerial vehicle to go from the warehouse to the customer points to take goods, and then returning to the warehouse; for the customer points which only send goods, arranging an unmanned aerial vehicle to send goods to the customer points before loading the goods from the warehouse, and after the goods are sent, selecting the unmanned aerial vehicle to go to the next customer point to take the goods and then return to the warehouse by the unmanned aerial vehicle or directly returning to the nearby warehouse by the unmanned aerial vehicle; for a customer site with simultaneous goods taking and delivering tasks, an unmanned aerial vehicle is arranged to deliver goods from a warehouse, take the goods at the customer site after unloading, and then return to the warehouse.

Further, the task allocation module divides tasks into three types according to the task types of the client points, stores the task points which only have goods taking requirements and do not have goods taking requirements into a PICKUP set, stores the task points which only have goods taking requirements and do not have goods taking requirements into a DROP set, and stores the task points which have goods taking requirements and do not have goods taking requirements into a PICK-DROP set; and after the task classification is finished, generating an initial task allocation scheme considering the geographic position by using a k-means algorithm for the task points in the PICKUP, the DROP and the PICK-DROP set.

Further, the neighborhood operators include 2-exchange, 3-exchange, 30% -exchange, Relocation, Other-Relocation, and 10% -Relocation.

Further, the 2-exchange operator operation method is as follows: randomly selecting two task points i, j (i, j belongs to Pickup) under a warehouse in a planned path, and interchanging the positions of the two client points i, j; the operation method of the 3-exchange operator comprises the following steps: randomly selecting three customer points i, j and k (i, j, k belongs to Pickup) under a warehouse in the planned path, and then randomly exchanging the positions of the three customer points i, j and k; the 30% -exchange operator is specifically operated as follows: randomly selecting 30% of the customer points in the planned path and then randomly disordering the order of the customer points.

Further, the Relocation operator specific operation method is as follows: randomly selecting a client point i (i belongs to PICKUP) in the warehouse a, randomly selecting an insertion point in the existing path planning scheme of the warehouse b, and migrating the client point i to the path planning scheme of the warehouse b according to the model constraint condition; the specific operation method of Other-relocation is as follows: selecting a client point i (i belongs to PICKUP U PICK-DROP) from all tasks distributed to the warehouse a, and randomly inserting the client point i into a path scheduling scheme of the warehouse b to be executed according to a model constraint condition; the specific operation method of the 10% -relocation comprises the following steps: and randomly selecting one warehouse each time, and migrating the client points farthest from the warehouse according to the distance between the client points and the warehouse, wherein the number of the migrated client points is 20% of the total client points of the warehouse.

Further, the path planning module generates an initial solution S₀The method comprises the following specific steps:

generating task allocation schemes of all warehouses: the tasks distributed to each warehouse are divided into three types of tasks of only taking goods, only delivering goods, simultaneously taking goods and delivering goods according to the characteristics of the tasks and respectively storedG_k-pick、 G_k-drop、G_k-pick&Among three sets of drop, initializing set

Storing the task sequence of the warehouse k; randomly fetch G_k-storing of task points in drop collections into P_kThen randomly take out G_k-task point deposit P in pick set_kFinally, G is_k-pick&Storing task points in drop set into P in sequence_k(ii) a Putting the task sequences of all the warehouses together to obtain a processed task allocation scheme;

and (3) path planning is carried out: when an initial solution is constructed, assigning task points in a DROP set and a PICK-DROP set to different unmanned aerial vehicles, and then randomly allocating the PICKUP type task points to the unmanned aerial vehicle which is already assigned with the DROP task or assigning a new unmanned aerial vehicle again for execution; and constructing a plurality of different path planning schemes based on the task allocation scheme, and selecting the scheme with the minimum cost as an initial solution according to a greedy principle.

Further, the unmanned aerial vehicle path scheduling scheme under each warehouse is re-planned according to the initial solution, and the specific steps are as follows:

given an initial solution s₀Selecting one from 6 neighborhood operators to perform neighborhood search on the initial solution to form a new solution

Calculating the cost of the new solution

If the cost of the new solution is less than the cost(s) of the initial solution₀) Then accept the new solution

Otherwise, accepting the new solution with a certain probability

At the same time handle

Is assigned to

If the two conditions are not satisfied, the initial solution s is solved₀Is assigned to

Will be provided with

Storing in solution space S, updating S₀And cost(s)₀) A value of (d);

repeating the steps until the loop termination condition is met, and then finding the path planning scheme with the minimum cost from the solution space S

As the current optimal scheduling scheme.

Further, randomly selecting a warehouse in the current optimal scheduling scheme, and performing local search on the scheduling scheme of the warehouse, specifically comprising the following steps:

classifying the client points to be executed by the selected warehouse: customer points of pick-only sort to s_1k-pick; delivery-only customer points categorized into s_1k-drop; customer site classification with simultaneous pick and delivery tasks to s_1k-pick&drop; when s is_1k-pick and s_1kWhen none of the drops is empty, will s_1k-sorting in pick the customer site where the drone departs from the warehouse to the customer site for picking and then returns to the warehouse into pick₁A 1 is to_1k-sorting of customer sites within a drop collection from a warehouse to a customer site for delivery and then back to the warehouse by an unmanned machine₁；

From pick₁And drop₁Randomly selecting one goods-taking customer point p in the two sets respectively₁Delivery-only customer site d₁And the two tasks originally assigned to the two unmanned aerial vehicles to be completed are combined and completed by one unmanned aerial vehicle.

The unmanned aerial vehicle goods taking and delivering method based on the urban roof automatic express device disclosed by the second aspect of the invention is applied to the system and comprises the following steps:

acquiring position information and goods taking and delivering information of multiple warehouses, multiple unmanned aerial vehicles and multiple customer points;

designing a comprehensive scheduling problem based on unmanned aerial vehicle parcel taking and delivering, multi-center unmanned aerial vehicle load and range, multi-customer-point unmanned aerial vehicle and multi-unmanned aerial vehicle;

and a task allocation stage: allocating an initial task to the single center, and reallocating the task according to operators in various fields;

a path planning stage: finding a path planning scheme for completing the task for each warehouse according to various neighborhood operators;

continuously iterating the task allocation stage and the path planning stage of the interaction step, and searching a global optimal solution until an algorithm termination condition is met; setting a small loop and a large loop, wherein the small loop generates a new scheduling scheme meeting constraint conditions, the large loop carries out local search on the basis of the scheduling scheme generated by the small loop to form a new scheduling scheme, and then determines whether to accept the result of the local search according to a greedy principle;

and distributing multiple unmanned aerial vehicles according to the global optimal solution to take and deliver goods from multiple warehouses and multiple customer points.

The invention has the following beneficial effects:

the vacant space of the city is fully utilized;

the straight-line flying can be carried out on the way, and the taking and delivering efficiency is high;

competition of space backlog and queuing with traditional goods taking and delivering is avoided;

compared with 6 heuristic and non-heuristic algorithms, the method generates a high-quality scheduling scheme within reasonable time, and is superior to other heuristic and meta-heuristic algorithms in the aspects of solving quality and time efficiency.

Drawings

FIG. 1 is a flow chart of a method for picking up and delivering goods by an unmanned aerial vehicle based on two-stage optimization and iteration;

FIG. 2 is a schematic diagram of the unmanned aerial vehicle pick-and-place system of the present invention;

FIG. 3 illustrates an encoding method according to the present invention;

FIG. 4 is a schematic view of the 2-exchange of the present invention;

FIG. 5 is a 10% -relocation scheme of the present invention;

FIG. 6 is a diagram illustrating the 10% -relocation of the present invention after allocation;

FIG. 7 is a schematic diagram of a partial search according to the present invention;

FIG. 8 is a graph comparing SATO-SVND revenue for various algorithms and the present invention;

FIG. 9 is a graph of the results of different algorithms;

FIG. 10 is a comparison of different algorithm run times;

FIG. 11 is a site map label for use with the present invention;

fig. 12 is an iterative droop graph of the present invention.

Detailed Description

The invention is further described with reference to the accompanying drawings, but the invention is not limited in any way and any alterations or substitutions based on the teaching of the invention are within the scope of the invention.

As shown in fig. 1, the unmanned aerial vehicle pick-and-place system based on two-stage optimization and iteration disclosed by the invention comprises the following steps:

the data input module acquires position information and goods taking and delivering information of multiple warehouses, multiple unmanned aerial vehicles and multiple customer points;

the design of the scheduling module is based on the comprehensive scheduling problems of unmanned aerial vehicle parcel taking and delivering, multi-center of unmanned aerial vehicle load and range, multi-customer point and multi-unmanned aerial vehicle;

a task allocation module: allocating an initial task to the single center, and reallocating the task according to operators in various fields;

a path planning module: finding a path planning scheme for completing the task for each warehouse according to various neighborhood operators;

an optimization module: continuously iterating an interaction task allocation stage and a path planning stage, and searching a global optimal solution until an algorithm termination condition is met; setting a small loop and a large loop, wherein the small loop generates a new scheduling scheme meeting constraint conditions, the large loop carries out local search on the basis of the scheduling scheme generated by the small loop to form a new scheduling scheme, and then determines whether to accept a result of the local search according to a greedy principle;

the goods taking and delivering module: and distributing multiple unmanned aerial vehicles to take and deliver goods from multiple warehouses and multiple customer points according to the global optimal solution.

The following sections specifically illustrate the principles of the present invention.

Hypothesis of problem

The last kilometer-oriented unmanned aerial vehicle goods taking and delivering system is composed of a plurality of unmanned aerial vehicles and a plurality of automatic storing and delivering devices. Suppose a plurality of unmanned aerial vehicle logistics centers are distributed in a city, each center is provided with a plurality of unmanned aerial vehicles, and each unmanned aerial vehicle logistics center can be used as a warehouse. The roof of the urban community is taken as delivery and pick-up points of goods, and each pick-up/delivery point is taken as a client point. The automatic storage equipment and the automatic goods loading and unloading device are configured on the roof, and the unmanned aerial vehicle can go to the next place after goods are loaded and unloaded in the warehouse. Meanwhile, the unmanned aerial vehicle can automatically change the battery on the roof, and the changed battery is charged on the roof, so that the conveying efficiency and the voyage of the unmanned aerial vehicle are improved.

As shown in fig. 2, the present invention has a plurality of customer sites, a plurality of warehouses, and a plurality of drones. In fig. 2, the drone has multiple routes, e.g., the drone may be loaded from a warehouse to a customer site, and after unloading at the customer site, the drone may choose to return to the nearest warehouse or go to the next customer site; the unmanned aerial vehicle can go from the warehouse to a customer point to pick up goods and then return to the warehouse; a drone may be loaded from a warehouse to a customer site, and if the customer site has the task of picking up goods, the drone unloads at the customer site, reloads the goods, and then returns to the warehouse. The problem can be regarded as a comprehensive scheduling problem which considers unmanned aerial vehicle parcel taking and delivering, multi-center of unmanned aerial vehicle load and range, multi-customer point and multi-unmanned aerial vehicle. In terms of obstacle avoidance, non-patent literature "human Liu, An automatic road Planning Method for Unmanned Aerial Vehicle Based on a targeted Intersection and targeted Guidance Strategy" proposes An Unmanned Aerial Vehicle automatic obstacle avoidance Method, which generates two paths between two points Based on Tangent Intersection and a targeted Guidance Strategy when An obstacle is encountered, and selects one Path according to a heuristic rule, so that a satisfactory collision-free Path can be generated in An uncertain environment in An approximately real-time manner. In the invention, the warehouse and the pick-and-place point are arranged on the roof of the city, and the unmanned aerial vehicle has less obstacles, so the invention mainly considers two aspects of task allocation and path planning, and the specific three-dimensional obstacle avoidance method can refer to the method. To the parcel pickup and delivery problem of multicenter, many customer points, many unmanned aerial vehicles, there is the following hypothesis:

(1) goods transported by unmanned aerial vehicle are packaged by special boxes with uniform size

(2) And after the battery of the unmanned aerial vehicle is replaced, the remaining voyage becomes the maximum voyage.

(3) Unmanned aerial vehicle directly flies to customer's point, can not midway change the battery and go to customer's point again by the detour.

(4) The drone can only take one package at a time.

(5) Unmanned aerial vehicle is in order flying at the uniform velocity, and does not consider the energy consumption that unmanned aerial vehicle took off and land.

Objective function

Description of the essential elements: depots set B: {1,2, …, m }, UAV set U: {1,2, …, k }, customer point set C ═ 1,2, …, C }. In order to describe that a client only takes goods, only delivers goods and takes delivery goods to carry out three different task types simultaneously, a set DROP of client points is defined as the client points of the delivery-only task; defining a set PICKUP of client points as client points only picking up goods; the set of customer points PICK-DROP is defined as customer points having both PICK-up and delivery tasks. In order to distinguish two different operations of picking and delivering goods, each point in the customer point set is divided into two points, one point is used for describing a goods picking task and the other point is used for describing a goods delivering task, namely C^pick＝{1,2,...,c}，C^dropAs { c +1, c +2,.., 2c }. For example, originally

Corresponding point i eC^pickAnd point i + C ∈ C^drop. h is the maximum range of the unmanned aerial vehicle when in no-load.

TABLE 1 symbol definitions

(symbol)	Description of the invention
		B	Warehouse set, B ═ {1,2, …, m }
U	Set of drones, U ═ 1,2, …, k }
		C	Set of customer points, C ═ {1,2, …, C }
i,j	Customer index
		k	Unmanned aerial vehicle index
n	Task indexing
		u_k	Flight cost of UAV k
T_i	Task type of customer Point i
		c_n	Weight of task n
c_max	Maximum capacity of unmanned aerial vehicle

Definition of the variables 0 to 1

If drone k performs a delivery/pickup task from customer point i to customer point j,

otherwise

Definition of d_i,jThe flight distance of the unmanned plane k from the client point i to the client point j; definition u_kStarting times for the unmanned aerial vehicle.

Defining an integer variable T_iIndicating the task type of the client point i. Wherein, T _i1 indicates that customer site i has delivery tasks, T _i0 denotes no delivery/pick-up task at customer site i, T _i1 indicates that customer site i has a pick up task.

Definition of

The maximum range when unmanned aerial vehicle k is loaded with cargo n. The flight time of the unmanned aerial vehicle is linearly decreased along with the increase of the effective load. Therefore, it can be assumed that there is a linear correlation relationship between the range of the drone and the load, and when the drone k flies from the point i to the point j, the maximum range of the drone decreases with the increase of the weight of the cargo carried by the drone at this time, which is specifically expressed as follows:

the size of the payload may affect the energy consumption and flight range of the drone. When the electric quantity of the unmanned aerial vehicle is constant, a linear correlation relation exists between the electric quantity attenuation and the load capacity of the unmanned aerial vehicle. The invention specifically expresses the cargo load penalty coefficient beta as follows:

wherein, beta_maxIs the maximum value of the penalty coefficient, c_maxThe maximum load capacity of the unmanned aerial vehicle is, and when the unmanned aerial vehicle is unloaded, beta is 1; when the unmanned aerial vehicle is fully loaded, beta is beta_max。

Optimization objective of the model: the first aim of the invention is to ensure that the total flight path for the drone to serve all customers and return to the warehouse is minimal; in addition, the model is an mD-PSTSP problem, and therefore, a second optimization goal of the model is to use as few drones as possible to complete the task, considering that each drone is started with corresponding energy consumption and depreciation. The objective of the objective function is to minimize the total energy consumption of the drones and the weighted sum of the total number of drones.

And alpha and rho are weight coefficients of the total travel distance of the unmanned aerial vehicle and the starting frame number of the unmanned aerial vehicle.

Constraint conditions are as follows:

c1 (decision variables):

c2 (task point constraints):

c3: the drone must start from the warehouse and finally return to the warehouse:

c4 (service point constraint):

after the unmanned aerial vehicle delivers the goods, the unmanned aerial vehicle can select the next goods taking point or warehouse; because the unmanned aerial vehicle can only carry one package at a time, the unmanned aerial vehicle must return to a warehouse immediately after taking goods; for optimal cost, a task point with simultaneous taking and delivering tasks is completed by an unmanned aerial vehicle:

c5 (range constraint):

when drone k is loaded with package n to fly from customer point i to customer point j, the maximum range of the drone cannot be exceeded.

Constraint 6:

and (3) load capacity constraint: the weight of goods taken and delivered by the unmanned aerial vehicle each time cannot exceed the maximum load of the unmanned aerial vehicle;

wherein, c_nThe weight of the goods for the nth customer point.

The problem of parcel taking and delivering of multiple centers, multiple customer sites and multiple unmanned aerial vehicles is a very complex combined optimization problem. The invention designs a two-stage optimization method as a solution to the problems so as to improve the computing power.

Algorithm framework

With the increase of customer points, the calculation complexity of the mD-PSTSP is increased sharply. Traditional heuristic algorithms and meta-heuristic algorithms have difficulty finding a high-quality path planning scheme in a short time. Therefore, the invention designs a two-stage optimization method which is named as SATO-SVND algorithm. The original multi-warehouse, multi-customer-point and multi-unmanned aerial vehicle task scheduling problem is decomposed into a plurality of simple single-center task scheduling problems, and then respective path planning schemes are independently solved. The algorithm consists of two stages: a task allocation stage for converting the multi-center task scheduling problem into a single-center task scheduling problem; and a path planning stage of the single center, namely a path planning scheme stage for finding a completed task for each warehouse. In order to find the global optimal solution, the two stages need to be continuously iterated and interacted.

The method comprises the steps of performing initial distribution of tasks in a first stage, dividing the tasks into three types according to task types of client points, storing task points with only goods taking requirements and without goods taking requirements into a PICKUP set, storing task points with only goods taking requirements and without goods taking requirements into a DROP set, and finally storing task points with goods taking requirements and with goods taking requirements into a PICK-DROP set. After task classification is completed, task points in PICKUP, DROP and PICK-DROP sets are generated into an initial task allocation scheme considering geographic positions by adopting a k-means algorithm, and a multi-center task scheduling problem is converted into a single-center task scheduling problem. At this time, the result of the task being distributed to each warehouse is an unordered state, and the task cannot be directly handed to the unmanned aerial vehicle to be executed. Therefore, we need to do path planning.

In the second stage, an SVND algorithm containing 6 kinds of neighborhood operators is designed according to the distribution result of each single-center task, and a path planning scheme meeting the constraint condition is generated. At the moment, all the central tasks are ordered, and all the central task scheduling schemes are combined together to form a complete task scheduling scheme which can be delivered to the unmanned aerial vehicle for execution. In order to adjust the existing scheme and obtain a better solution, the invention designs 6 operators (2-exchange, 3-exchange, 30% -exchange, relocation, other-relocation, 10% -relocation). 3 operators such as 2-exchange and 3-exchange, 30% -exchange and the like are used for adjusting a task allocation scheme in one warehouse, and tasks are exchanged between two or three unmanned aerial vehicles under one warehouse. relocation and other-relocation, 10% -relocation are the task allocation scheme for adjusting two warehouses, and the task allocated to one warehouse point is transferred to the other warehouse.

The two phases are iterated until the algorithm termination condition is met. Two loop structures are set in the algorithm: the small loop uses a variable neighborhood search algorithm to generate a new scheduling scheme that satisfies the constraint. Judging whether to accept the new scheduling scheme or not according to a metropolis principle every time, if so, recording the new scheduling scheme until the number of the new scheduling scheme meets a preset number, and forming a solution space by the recorded new scheduling scheme; the large circulation is to search an optimal scheduling scheme on the basis of a solution space generated by the small circulation, and the temperature is reduced every time circulation is carried out until the temperature reaches the lowest temperature. In order to jump out the Local optimum, a Local Search algorithm is designed in the large loop, the current best solution found in the solution space at each time needs to be subjected to Local Search again, the existing solution structure is destroyed through a Local Search algorithm to form a new scheduling scheme, and whether a Local Search result is accepted or not is determined according to a greedy principle. The two-stage algorithm successfully converts the multi-center mD-PSTSP into the single-center mD-PSTSP, and the difficulty in solving the original problem is reduced.

TABLE 2 SATO-SVND Algorithm pseudocode

Coding method

The scheduling scheme under each warehouse is specified to go from the warehouse to a task point and finally return to the warehouse; the unmanned aerial vehicle can directly return to the warehouse after starting from the warehouse to a task point of delivery or return to the warehouse after going to a nearest delivery point to take a delivery; because the unmanned aerial vehicle can only carry a parcel at a time, the unmanned aerial vehicle must return to the warehouse after getting the goods. In summary, the path of the drone has the following three conditions: 1) the unmanned aerial vehicle goes from the warehouse to a customer point to pick up goods and then returns to the warehouse; 2) the unmanned aerial vehicle goes from the warehouse to a customer point to deliver goods and then returns to the warehouse; 3) the unmanned aerial vehicle goes to the customer point from the warehouse to deliver goods, and then returns to the warehouse after getting goods at the customer point.

In the prior art, when solving a path problem, the heuristic algorithm and the meta-heuristic algorithm usually do not consider warehouse points at first, only encode client points to be accessed, decode the path under the condition that a complete path is needed for subsequent distance cost calculation, adaptability and the like, and add the warehouse points to form the complete path. The invention adopts a new coding mode in consideration of the need of distinguishing three different situations, namely, the situation that goods are returned to the warehouse after being taken from the warehouse to the client point, the situation that goods are returned to the warehouse after being delivered from the warehouse to the client point and then taken from the client point, and the like. As shown in fig. 3, each warehouse and customer site has a unique number according to the strict constraint condition during coding, and is used for distinguishing the warehouse, the customer site with only goods taking requirement, the customer site with only delivery requirement and the customer site with both goods taking and delivery requirement. The numbering of the warehouse is continued on the basis of the numbering of the customer points, and if the last customer point is numbered x and the total number of the warehouse is m, the numbering of the warehouse points is x +1, x + 2. The merging of the scheduling schemes of the individual warehouses is a complete scheduling scheme.

Task allocation

(1) Task classification

Different from the problem of parallel unmanned aerial vehicle dispatching station traveler (PDSSP), the construction of a solution of the problem of Multi-warehouse and Multi-unmanned aerial vehicle parallel dispatching station traveler (m-Drone-PSTSP with drop and pick-up synchronization at Multi-drop) which is carried out when goods are fetched and delivered simultaneously needs to distinguish task types

(2) Initial task allocation

In the initial task allocation stage, a k-means clustering algorithm considering the geographic position is used for allocating the customer points to different warehouses, and the problem is converted into a single-center problem. Then, a single warehouse plans an unmanned aerial vehicle path, and assigns an unmanned aerial vehicle to visit each customer point to complete the goods taking and delivering task.

(3) Task reallocation

In order to optimize an initial task allocation result and find an optimal scheduling scheme, 6 neighborhood operators such as 2-exchange, 3-exchange, 30% -exchange, relocation and, other-relocation, 10% -relocation and the like are designed to redistribute the task allocation scheme.

(ii) 2-exchange: and exchanging the pick-up tasks distributed to the two unmanned aerial vehicles under the same warehouse. The specific operation method comprises the following steps: and randomly selecting two task points i, j (i, j belongs to Pickup) under a warehouse in the planned path, and exchanging the positions of the two client points i, j. The specific operation is shown in fig. 4.

② 3-exchange: and exchanging delivery tasks distributed to three unmanned aerial vehicles under the same warehouse. The specific operation method is similar to the 2-exchange operation: and randomly selecting three customer points i, j and k (i, j, k belongs to Pickup) under a warehouse in the planned path, and then randomly exchanging the positions of the three points i, j and k.

(iii) 30% -exchange: in the face of large-scale tasks, the optimization efficiency of 2-exchange and 3-exchange is low, so that 30% -exchange is developed. When the warehouse services have more than 10 customer points, we use 30% -exchange, which exchanges the order of 30% of the customer point accesses under a single warehouse at a time. The specific operation is similar to 2-exchange: randomly selecting 30% of the customer points in the planned path and then randomly disordering the order of the customer points.

(iv) Relocation: randomly selecting one picking task from all the tasks distributed to the warehouse a to be transferred to the warehouse b for execution. The specific operation method comprises the following steps: randomly selecting a client point i (i belongs to PICKUP) in the warehouse a, randomly selecting an insertion point in the existing path planning scheme of the warehouse b, and migrating the client point i to the path planning scheme of the warehouse b according to the model constraint condition. As shown in fig. 5 and 6.

Other-relocation: and selecting one delivery task which is farthest away from the warehouse from all tasks distributed to the warehouse a to be transferred to the warehouse b for execution. The specific operation method comprises the following steps: and selecting a client point i (i belongs to PICKUP U PICK-DROP) in all tasks distributed to the warehouse a, and randomly inserting the client point i into the path scheduling scheme of the warehouse b for execution according to the model constraint condition.

Sixthly, 10% -relocation: similar to 30% -exchange, we developed 10% -relocation to handle the case of large-scale tasks. When the warehouse service has more than 20 customer sites, we use 10% -relocation. The specific operation method comprises the following steps: and randomly selecting one warehouse every time, and migrating the customer points farthest from the warehouse according to the distance between the customer points and the warehouse, wherein the number of the migrated customer points is 20% of the total customer points of the warehouse.

Path planning

(1) Generating an initial solution

For the convenience of solution, for a customer site with multiple pick/deliver package tasks, the customer site may be split into multiple customer sites, each having at most one pick and one delivery task. For customer sites with simultaneous pick-up and delivery requirements, pick-up and delivery tasks for such customer sites are specified to be performed by the same drone. Because the unmanned aerial vehicle can only load one package at a time, when the packages of the unmanned aerial vehicles are distributed in parallel, the task points in the DROP and PICK-DROP sets all need the independent unmanned aerial vehicles to load goods from the warehouse to be delivered to the client; the task points in the PICK-DROP set have delivery tasks and PICK-up tasks at the same time, so that the unmanned aerial vehicle can directly PICK up goods in situ after delivering the goods in the PICK-DROP set; the unmanned aerial vehicle can start from a warehouse to load goods; after the unmanned aerial vehicle delivers the goods, the goods can be directly returned to a warehouse or go to the next goods taking task point to take the goods according to the requirement; all drones must return to the warehouse for unloading after loading the goods. The above is the path planning principle of the warehouse.

After the tasks are distributed, the tasks are in an unordered state, and the tasks are sequenced to form an initial solution S₀. For the task distributed to each warehouse, firstly, the task is divided into three types of only goods taking, only goods delivery, simultaneously goods taking and goods delivery tasks and the like according to the characteristics of the task, and the three types of tasks are respectively stored in G_k-pick、G_k-drop、G_k-pick&drop three sets. Initializing a set

The task sequence of warehouse k is stored. Randomly fetch G_k-storing of task points in drop collections into P_kThen randomly take out G_k-task point deposit P in pick set_kFinally, G is_k-pick&Storing task points in drop set into P in sequence_k. Putting the task sequences of all the warehouses together is a processed task allocation scheme. After the scheme is obtained, path planning is carried out according to the constraint conditions 3-4. When constructing an initial solution, assigning task points in a DROP set and a PICK-DROP set to different unmanned aerial vehicles, and then randomly allocating task points of a PICKUP type to the unmanned aerial vehicles which are already allocated with DROP tasksOr reassign a new drone to execute. In order to guarantee the stability of the solution, a plurality of different path planning schemes are constructed based on a task allocation scheme, and a scheme with the minimum cost is selected as an initial solution based on a greedy principle.

(2) Iterative optimization

Based on a two-stage optimization framework, a simulated annealing Algorithm is provided, tasks are redistributed by a plurality of conversion factors, and after the tasks are redistributed, an unmanned aerial vehicle path scheduling scheme under each warehouse is re-planned by a Variable Neighborhood search Algorithm (SVND).

The variable neighborhood search algorithm is an improved meta-heuristic optimization algorithm and can be used for solving a combined optimization problem. The invention designs a special variable neighborhood search algorithm: on condition that the algorithm loop is satisfied, I give it an initial solution s₀Then, the initial solution is optimized by using a neighborhood structure formed by different actions, and a new solution can be obtained in each circulation. The new solution generated is to be compared with the initial solution s₀And comparing, judging whether the new solution is good or bad and determining whether to accept the new solution. The valid solutions are stored in a set of solutions PICK-DROP. And repeating the steps until the maximum iteration number is reached. The optimal solution in the solution set S is the optimal path planning scheme searched by the current algorithm. It is worth noting that in order to jump out the local optimum and obtain the global optimum solution, we use the Metropolis principle to judge whether to accept the new solution.

For the SVND algorithm, we give it an initial solution s₀Selecting one from 6 neighborhood operators such as 2-exchange, 3-exchange, 30% -exchange, relocation and, other-relocation, 10% -relocation to perform neighborhood search on the initial solution to form a new solution

Calculating the cost of the new solution

If the cost of the new solution is less than cost(s)₀) Then accept the new solution

Otherwise, accepting the new solution with a certain probability

Handle at the same time

Is assigned to

If the above two conditions are not satisfied, then s is set₀Is assigned to

Will be provided with

Storing in the solution space S, updating S₀And cost(s)₀) The value of (c). And repeating the steps until the cycle termination condition is met. Then finding the path planning scheme with the minimum cost from the solution space S

As the output of the SVND algorithm. Table 3 gives the pseudo code for SVND.

TABLE 3 SVND pseudocode

Local search

In order to jump out of the local optimization of the variable neighborhood algorithm and obtain a global optimal solution, a local search algorithm is designed, namely a current optimal scheduling scheme is found in a solution space generated by the SVND algorithm, and local search is carried out on the basis of the optimal scheme.

The specific operation method is that a warehouse is randomly selected from the currently found optimal scheduling schemes, and the scheduling scheme of the warehouse is locally searched. Firstly, classifying the client points required to be executed by a selected warehouse: customer points of pick-only sort to s_1k-pick; delivery-only customer points categorized into s_1k-drop; customer site classification with simultaneous pick and delivery tasks to s_1k-pick&And (4) drop. When s is_1k-pick and s_1kWhen none of the drops is empty, we will s_1k-sorting into picks the customer sites within the picks where the drone departs from the warehouse to the customer site for pick-up and then returns to the warehouse₁In the same way, will s_1k-sorting into drop the customer points within a drop collection where drones leave the warehouse for delivery to the customer point and then return to the warehouse₁. From pick₁And drop₁Randomly selecting one goods-taking customer point p in the two sets respectively₁Delivery-only customer site d₁Combining the two tasks originally assigned to two unmanned aerial vehicles to complete, and completing by one unmanned aerial vehicle; that is, the drone departs from the warehouse d₁Delivery of goods, followed by p-removal₁The goods are taken and then returned to the warehouse. Figure 7 shows the algorithm optimization process. Table 4 gives the pseudo code for the local search.

TABLE 4 partial search pseudocode

Simulation experiment

In this part, the effectiveness of the invention is evaluated by comparing with 6 heuristic algorithms and metaheuristic algorithms, and experiments are carried out in a real scene. The algorithm and other seven algorithms provided by the invention are operated on a PC (personal computer) which utilizes python to carry out coding, wherein the CPU is Core i5-8400, the memory is 8G, and the operating system is Windows 10.

Experimental setup

In the problems studied by the invention, the problems of customer scale, warehouse quantity, unmanned aerial vehicle load, association between unmanned aerial vehicle load and voyage, customer distribution and the like need to be considered. Because the PDSTRP problem associated with mD-PSTSP is short in study time, there is no uniform test data set for reference for the moment. Therefore, when the effectiveness of the algorithm is tested, a randomly generated data set and 6 heuristic and meta-heuristic methods are selected to be used for experiments to prove the effectiveness of the algorithm.

TABLE 5 Experimental Algorithm parameter Table

The simulated scene is suitable for a region of 50KM × 50 KM. The customer points are randomly distributed within the area. The pick-and-deliver demand for each customer site is randomly generated. The weight of each pick-and-send package is randomly generated, assuming all drones are performing the same, with the weight of each package being between 1KG and 8 KG. The relevant parameter settings of the performance parameter setting set SATO-SVND algorithm of the unmanned aerial vehicle are shown in Table 6. We generated 6 examples 40, 60, 80, 100, 150, 200, and the parameter settings of the algorithm of the present invention were finally determined based on relevant literature and experimental trial and error.

Comparison with other meta-heuristic methods and heuristic methods

To verify the performance of the SATO-SVND algorithm, we compared the SATO-SVND algorithm with seven other heuristics and meta-heuristics. We set up 13 cases and the number of customer points is set to 40, 60, 80, 100, 150, 200. Each algorithm was run 10 times to solve each case and finally compared with the average of 10 experimental results, and the data in the table are the average of 10 experimental results. The experiment of the invention is provided with two reasons: firstly, the advantages of the two-stage algorithm in solving the goods taking and delivering problems of multiple warehouses, multiple customer points and multiple unmanned aerial vehicles are researched, and then the generation of the initial solution of the SATO-SVND algorithm and the effectiveness of the neighborhood structure are considered.

Aiming at the problems of task allocation and path planning of the unmanned aerial vehicle, the advantages of a two-stage algorithm in solving the goods taking and delivering problems of multiple warehouses, multiple customer points and multiple unmanned aerial vehicles are verified, and three high-efficiency algorithms, namely ALNS, k-means & GA and k-means & LNS algorithms are designed as comparison algorithms. As a comparison algorithm, we used a two-stage solution framework and tournament operators to develop the k-means & GA algorithm. Meanwhile, in order to ensure that good individuals of the parents are not lost due to crossover or mutation operations, an elite strategy is adopted to keep the optimal individuals of the parents. The ALNS algorithm and the k-means & LNS algorithm adopt a damage repair operator, and repeatedly iterate optimization under the framework of simulating a return algorithm until a termination condition is met. To accommodate the solution of large-scale problems, we destroy 10% -20% of the solution structure at a time, and then repair the solution with random and greedy repair operators. In K-means & LNS, a task is firstly distributed to a plurality of warehouses by using a K-means algorithm, and then neighborhood structures such as 2-opt, simple relocation, and swap move and the like are adopted for searching in path planning. The neighborhood structure of the ALNS is then merged into an adaptive mechanism and a worst removal operator.

In order to prove the correctness of the solution thought of the two-stage algorithm and the effectiveness of the neighborhood structure, three algorithms of SA-SVND and ETSA adopting random allocation tasks and SATO-VND1 removing local search are compared. In order to verify the effectiveness of the two stages, the SA-SVND adopts a random distribution mode in a task distribution stage, then the SVND algorithm designed by the invention is adopted in a single-center unmanned plane path planning stage, iteration is repeated under the framework of the SA, and finally the optimal solution is found. SATO-VND1 is used to remove the local search algorithm from SATO-SVND to verify the validity of the local search algorithm. These 6 algorithms were run 10 times for each instance. The calculation results are shown in fig. 8.

To better compare the differences between the algorithms, we used the Gap index to describe the differences between SATO-SVND and several other algorithms:

wherein S_iRepresents the optimal solution of the algorithm i, i represents the algorithm ALNS, k-means&GA、k-means&One of LNS, SA-SVND, ETSA, and SATO-VND 1; s_SATO-SVNDRepresents the optimal solution of the SATO-SVND algorithm of the invention. The calculation results are shown in table 7.

From the cost point of view, as shown in fig. 8, the SATO-SVND algorithm designed by us has absolute advantages in the solution of 13 cases compared with other 6 heuristics and meta-heuristics. For better comparison of the merits and demerits of the algorithm, we process the experimental results. The optimal cost obtained by solving each case by using other seven heuristic algorithms and the meta-heuristic algorithm is compared with the optimal cost obtained by solving the same case by using the SATO-SVND algorithm, and the difference between the optimal solutions obtained by using the other seven algorithms and the SATO-SVND is calculated. As can be seen from FIG. 8, the SATO-SVND algorithm always outperforms several other algorithms as customer sites increase. We can notice that the cost suddenly increases when calculating the two cases c6 and c13, which is caused by the large increase of customer points. We can see that SATO-SVND is very close to the optimal solution obtained by other algorithms when processing small-scale problems, but when processing large-scale problems, the effects of other algorithms are reduced to different degrees, and the difference between the optimal solution obtained by the ALNS algorithm and the optimal solution obtained by SATO-SVND calculation even reaches more than 30%.

TABLE 6 optimal solution of different algorithms and Gap of optimal solution of algorithm of the present invention

The gap between SATO-SVND and SATO-VND1 arises because SATO-VND1 does not use the Local Search operator to cause convergence and a locally optimal solution. However, SATO-SVND is very similar to SATO-VND1 in terms of computation time, indicating that Local Search can improve the solution in very little time consumption.

SATO-SVND is obviously superior to SA-SVND algorithm in cost, but the calculation time of the two algorithms is very similar, which shows that the SATO-SVND algorithm can produce better solution in reasonable time compared with the SA-SVND algorithm.

ETSA is obviously superior to the SATO-SVND algorithm in the calculation time, but the cost of a solution scheme obtained by the SATO-SVND algorithm is obviously superior to that of the ETSA algorithm, so that the quality of the solution scheme can be improved to a great extent by the SVND algorithm, the ETSA algorithm is easy to converge on a local optimal solution, and the ETSA algorithm is only suitable for the situation that the time is very sensitive and the requirement on the quality of the solution scheme is low.

We plot the boxed graphs with the calculation results of example 8, and from FIG. 9, it can be seen that the SATO-SVND algorithm is significantly better than the other six comparison algorithms, regardless of the stability and optimization effect of the algorithm.

From the time efficiency point of view, ETSA is superior to other seven algorithms, and the three algorithms of SA-SVND, SATO-VND1 and the like are the second order. In contrast, as customer sites increase, the time consumed by search algorithms such as k-means & GA, k-means-LNS, ALNS, etc. increases rapidly. Among them, k-means & GA performed the worst in time efficiency, followed by k-means-LNS, ALNS. The difference between the k-means-LNS algorithm and the ALNS algorithm is that the k-means-LNS algorithm generates an initial solution based on a two-stage optimization method, while the ALNS algorithm generates the initial solution randomly, and from the comparison of the two algorithms, the two-stage optimization method can greatly improve the optimization result of the algorithm within smaller time consumption, but the k-means & LNS still has the problem of premature convergence, and the SATO-SVND algorithm is far beyond the k-means & LNS in the aspect of cost optimization due to the adoption of various variable neighborhood structures. Meanwhile, SATO-SVND computation time is significantly shorter than k-means & LNS, indicating that SATO-SVND can produce better scheduling schemes in a shorter time than SATO-LNS. The algorithm runtime results are shown in fig. 10.

In order to further verify the effectiveness of the SATO-SVND algorithm in solving the large-scale task scheduling problem of multiple unmanned aerial vehicles, multiple warehouses and multiple client points, the client points under a real scene are selected as test data for further experimental verification. The real geographical coordinates of 80 buildings in 9 cells of the Yuenu area of Changsha in Hunan province are selected as the client points of the multiple unmanned aerial vehicles, and the specific positions are shown in FIG. 9. 5 warehouse points are arranged, and the weight of each cargo is a random number within 1KG-8 KG. The cost convergence curve is shown in fig. 11. The run time was 12.91s, and the optimal cost was 15.8% lower than the initial solution.

As can be seen from FIG. 11, the cost of the SATO-SVND algorithm solution scheduling scheme decreases at a fast speed within 50 iterations, which shows that the SATO-SVND algorithm has strong decreasing capability in a short time. The Metropolis principle, the related neighborhood structure and the local search operator avoid premature convergence of the algorithm. The algorithm converged to the optimal solution at 225 times.

The invention discloses an unmanned aerial vehicle goods taking and delivering method based on two-stage optimization and iteration, which comprises the following steps:

in the task allocation stage, an initial task is allocated to the single center, and task redistribution is carried out according to operators in various fields;

in the path planning stage, a path planning scheme for completing tasks is found for each warehouse according to multiple neighborhood operators;

continuously iterating a task allocation stage and a path planning stage, and searching a global optimal solution until an algorithm termination condition is met; setting a small loop and a large loop, wherein the small loop generates a new scheduling scheme meeting constraint conditions, the large loop carries out local search on the basis of the scheduling scheme generated by the small loop to form a new scheduling scheme, and then determines whether to accept a result of the local search according to a greedy principle;

The invention has the following beneficial effects:

the vacant space of the city is fully utilized;

compared with 6 heuristic and non-heuristic algorithms, the invention generates a high-quality scheduling scheme in reasonable time, and is superior to other heuristic and meta-heuristic algorithms in the aspects of solving quality and time efficiency.

The above embodiment is an embodiment of the present invention, but the embodiment of the present invention is not limited by the above embodiment, and any other changes, modifications, substitutions, combinations, and simplifications which do not depart from the spirit and principle of the present invention should be regarded as equivalent replacements within the protection scope of the present invention.

Claims

1. Unmanned aerial vehicle gets delivery system based on automatic express device in city roof, its characterized in that, automatic express device sets up in city roof, and the system includes:

2. The unmanned aerial vehicle pick-and-place system based on the urban roof automatic express device is characterized in that the unmanned aerial vehicle goes to the next customer point or warehouse after loading and unloading goods, or automatically changes batteries on the roof; for the customer points which only take goods, arranging an unmanned aerial vehicle to go from the warehouse to the customer points to take goods, and then returning to the warehouse; for the customer points which only send goods, arranging an unmanned aerial vehicle to start from the warehouse to load goods and send the goods to the customer points, and after the goods are sent, selecting the unmanned aerial vehicle to go to the next customer point to take the goods and then return to the warehouse by the unmanned aerial vehicle or directly returning to the nearby warehouse by the unmanned aerial vehicle; for a customer site with simultaneous goods taking and delivering tasks, an unmanned aerial vehicle is arranged to deliver goods from a warehouse, take the goods at the customer site after unloading, and then return to the warehouse.

3. The unmanned aerial vehicle PICK-and-place system based on the urban roof automatic express device according to claim 2, wherein the task allocation module classifies tasks into three categories according to task types of client points, stores task points with no PICK-and-place demand only in a PICKUP set, stores task points with no PICK-and-place demand only in a DROP set, and stores task points with PICK-and-place demand in a PICK-DROP set; and after the task classification is finished, generating an initial task allocation scheme considering the geographic position by using a k-means algorithm for the task points in the PICKUP, the DROP and the PICK-DROP set.

4. The unmanned aerial vehicle pickup system based on city roof automatic express delivery device of claim 1, wherein the neighborhood operators comprise 2-exchange, 3-exchange, 30% -exchange, Relocation, Other-Relocation and 10% -Relocation.

5. The unmanned aerial vehicle pick-and-place system based on the automatic express delivery device on the urban roof as claimed in claim 4, wherein the 2-exchange operator operation method is as follows: randomly selecting two task points i, j (i, j belongs to Pickup) under a warehouse in a planned path, and exchanging the positions of the two client points i, j; the operation method of the 3-exchange operator comprises the following steps: randomly selecting three customer points i, j and k (i, j, k belongs to Pickup) under a warehouse in the planned path, and then randomly exchanging the positions of the three customer points i, j and k; the 30% -exchange operator is specifically operated as follows: randomly selecting 30% of the customer points in the planned path and then randomly disordering the order of the customer points.

6. The unmanned aerial vehicle pick-and-place system based on the automatic express delivery device on the urban roof, according to claim 4, wherein the specific operation method of the Relocation operator is as follows: randomly selecting a client point i (i belongs to PICKUP) in the warehouse a, randomly selecting an insertion point in the existing path planning scheme of the warehouse b, and migrating the client point i to the path planning scheme of the warehouse b according to the model constraint condition; the specific operation method of Other-relocation is as follows: selecting a client point i (i belongs to PICKUP U PICK-DROP) from all tasks distributed to the warehouse a, and randomly inserting the client point i into the path scheduling scheme of the warehouse b to be executed according to the model constraint condition; the specific operation method of the 10% -relocation comprises the following steps: and randomly selecting one warehouse every time, and migrating the client points farthest from the warehouse according to the distance between the client points and the warehouse, wherein the number of the migrated client points is 20% of the total client points of the warehouse.

7. The unmanned aerial vehicle pick-and-place system based on automatic urban roof express delivery device according to claim 1, wherein the path planning module generates an initial solution S₀The method comprises the following specific steps:

generating task allocation schemes of all warehouses: the tasks distributed to each warehouse are divided into only goods taking, only goods delivering and only goods delivering according to the characteristics of the tasks,Simultaneously has three types of tasks of taking and delivering goods, and respectively stores in G_k-pick、G_k-drop、G_k-pick&Among three sets of drop, initializing set

8. The unmanned aerial vehicle pick-and-place system based on the automatic urban roof express delivery device according to claim 7, wherein the unmanned aerial vehicle path scheduling scheme under each warehouse is re-planned according to an initial solution, and the method comprises the following specific steps:

Calculating the cost of the new solution

Otherwise, accepting the new solution with a certain probability

At the same time handle

Is assigned to

Will be provided with

Storing in solution space S, updating S₀And cost(s)₀) A value of (d);

As the current optimal scheduling scheme.

9. The unmanned aerial vehicle pick-and-place system based on the urban roof automatic express delivery device according to claim 8, wherein a warehouse is randomly selected in the current optimal scheduling scheme, and the scheduling scheme of the warehouse is locally searched, specifically comprising the following steps:

classifying the client points to be executed by the selected warehouse: customer points of pick-only sort to s_1k-pick; delivery-only customer points categorized into s_1k-drop; customer site classification with simultaneous pick and delivery tasks to s_1k-pick&drop; when s is_1k-pick and s_1kWhen none of the drops is empty, will s_1k-unmanned aerial vehicle in pick goes from warehouse to customer site to pick up goods and then returns to warehouseClient points of a library are categorized into picks₁A 1 is to_1k-sorting into drop the customer points within a drop collection where drones leave the warehouse for delivery to the customer point and then return to the warehouse₁；

10. Unmanned aerial vehicle pick-and-place method based on city roof automatic express device, characterized in that, applied to the system of any one of claims 1-9, the method comprises the following steps:

designing a comprehensive scheduling problem based on unmanned aerial vehicle parcel pickup and delivery, multi-center of unmanned aerial vehicle load and range, multi-customer point and multi-unmanned aerial vehicle;

a path planning stage: finding a path planning scheme for completing the task for each warehouse according to a plurality of neighborhood operators;

continuously iterating the task allocation stage and the path planning stage of the interaction step, and searching a global optimal solution until an algorithm termination condition is met; setting a small loop and a large loop, wherein the small loop generates a new scheduling scheme meeting constraint conditions, the large loop carries out local search on the basis of the scheduling scheme generated by the small loop to form a new scheduling scheme, and then determines whether to accept a result of the local search according to a greedy principle;