A heterogeneous pesticide spraying robot task allocation and scheduling method

Abstract

The present application relates to a kind of heterogeneous pesticide spraying robot task allocation and scheduling method.The method obtains the characteristic data and task data of pesticide spraying robot, on the basis of considering pesticide spraying time window and the performance of heterogeneous pesticide spraying robot, task allocation and scheduling are carried out using adaptive greedy iterative algorithm, first, an initial solution is generated by efficient heuristic method, while using adaptive destruction mechanism, and according to the evolution state, destruction operator and destruction strength are automatically matched, then the solution is destroyed, the solution after destruction is regenerated in the reconstruction phase, and the solution is further optimized by combining local search optimization method, finally, when reaching the stopping time of algorithm, the best solution of heterogeneous pesticide spraying robot task allocation and scheduling is obtained.Compared with the prior art, the present application has the advantages of being able to complete the allocation and scheduling of pesticide spraying task efficiently in the appropriate pesticide spraying time interval, taking into account the different performance of heterogeneous robots, etc.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of multi-robot task allocation and scheduling, and particularly to a heterogeneous pesticide spraying robot task allocation and scheduling method. BACKGROUND

[0002] Currently, the world's agriculture is changing towards a "new type of agriculture" of globalization, standardization, high innovation, high value and high efficiency. Under the background of the rapid development of intelligent and unmanned technology, smart agriculture is facing new opportunities and challenges. Smart agriculture is crucial to solving the problem of labor reduction caused by population aging and the problem of low quality of agricultural products caused by various complex factors. In particular, the introduction of agricultural robots greatly improves the agricultural operation capacity, reduces the consumption of pesticides and fertilizers, and greatly reduces the workload of workers. In this process, the correct configuration of agricultural robot resources and the reasonable scheduling of tasks can further increase income, improve efficiency and reduce the input required in the agricultural process. Weeding is a tough job, and most farmers are willing to hand it over to robots to complete the weeding and pest control task through the cooperation of multiple pesticide spraying robots in the farmland.

[0003] The existing research on agricultural robot task allocation and scheduling faces some problems. First, the influence of pesticide spraying time on the efficiency and cost of agricultural robots executing tasks is not considered. Crops are vulnerable to weeds and pests during the seedling stage, which will compete with crops for water, fertilizer and sunlight, leading to poor nutrition of crops. Therefore, weeding and pest control is crucial to improving yield. Common weeding and pest control agents include alachlor, isopropylamine, atrazine, propazine, benzofluor, and thiazopyr. These agents have strict requirements for weather conditions such as temperature. High or low temperature conditions will affect the effectiveness of weeding and pest control agents. For example, in the middle of the day, the light is strong, the agent is easy to decompose and volatile, so the effect is poor. Therefore, in general, the best time for crop pesticide spraying is from 9 am to 11 am and from 4 pm to 7 pm. At about 9 am, the dew is dry, so the pesticide will not be diluted by the dew, and the pesticide will not decompose due to high temperature, reducing its effectiveness. After 4 pm, the sun is westward and the light is gradually weak, and the night-active pests will start to move, so spraying pesticides can effectively kill night-active pests. Second, the influence of different working efficiencies and capabilities of different types of robots, i.e. heterogeneous robots, is not considered. In the actual environment, due to technical differences, there are often multiple types of pesticide spraying robots in the agricultural machinery center, and their pesticide spraying efficiency, travel speed, coverage range and pesticide loading capacity may differ.

[0004] Therefore, how to improve the rationality of heterogeneous agricultural robot pesticide spraying task planning, and thus improve the pesticide spraying effect, save the pesticide spraying cost, reduce the execution cost of the robot system, improve the agricultural production efficiency and economic benefit, has become a problem to be solved in the field. SUMMARY

[0005] The present application aims to overcome the defects of the prior art, i.e., the poor rationality of the task planning of the heterogeneous robots, leading to poor spraying effect and high cost, and provides a task allocation and scheduling method for heterogeneous agricultural robots, which can efficiently allocate and schedule the spraying tasks in a suitable spraying time interval and take into account the different performances of the heterogeneous robots.

[0006] The object of the present application can be achieved by the following technical solutions:

[0007] The present application provides a task allocation and scheduling method for heterogeneous spraying robots, comprising the following steps:

[0008] A data acquisition step is used to acquire spraying robot characteristic data and task data, and to initialize the population size, the lower bound of the destruction strength, the upper bound of the destruction strength and the stopping time, wherein the task data comprises a spraying time window, and the destruction strength is the number of removed tasks;

[0009] A population initialization step is used to generate multiple initial solutions by using a heuristic method and a random method respectively and to calculate the corresponding comprehensive cost values, to form an initial population, and to select the initial solution with the minimum comprehensive cost value from the initial population, wherein the heuristic method comprises allocating tasks based on the spraying robot characteristic data and the task data, and selecting the best insertion position for each task based on the spraying time window and the comprehensive cost value, and the comprehensive cost value is the sum of the spraying robot delay penalty cost, the travel cost and the dispatching cost;

[0010] An adaptive destruction step is used to match the destruction operator and the initial value of the destruction strength according to the current evolution state, to perform a destruction operation according to the destruction operator and the current destruction strength, to acquire the current remaining task set and the destruction task set, and the destruction operator is related to the characteristic data or the task data of different spraying robots;

[0011] A solution reconstruction step is used to allocate all the tasks in the current destruction task set to the position with the minimum comprehensive cost value in the remaining task set in sequence, to obtain the current solution;

[0012] A local search optimization step is used to re-allocate all the tasks of each spraying robot in sequence by local search, to acquire the solution with the minimum comprehensive cost value;

[0013] A best solution acquisition step is used to determine whether the stopping time is reached, if not, the adaptive destruction step is returned, and if yes, the current solution is outputted, to obtain the best task allocation and scheduling scheme for the heterogeneous spraying robots.

[0014] As a preferred technical solution, the heuristic method comprises a fastest arrival heuristic method, and execution steps of the fastest arrival heuristic method comprise:

[0015] a time of arrival sorting step, distances of all task point positions from a departure position of the pesticide spraying robot and arrival times of the pesticide spraying robot are calculated, all tasks are sorted in ascending order of the arrival times, the pesticide spraying robot feature data comprises the departure position of the pesticide spraying robot and a moving speed, and the task data comprises the task point position, and the arrival times are calculated by using the distances and the moving speed;

[0016] a task allocation step, each task is allocated to a corresponding optimal insertion position based on the pesticide spraying time window and the comprehensive cost value until all task allocation ends.

[0017] As a preferred technical solution, the heuristic method comprises a fastest completion heuristic method, and execution steps of the fastest completion heuristic method comprise:

[0018] a pesticide spraying time sorting step, pesticide spraying times of all tasks are calculated, and all tasks are sorted in ascending order of the pesticide spraying times, the task data comprises a pesticide demand of a task point, and the pesticide spraying times are determined according to the pesticide demand of the task point;

[0019] a task allocation step, each task is allocated to a corresponding optimal insertion position based on the pesticide spraying time window and the comprehensive cost value until all task allocation ends.

[0020] As a preferred technical solution, the heuristic method comprises a Max-Max heuristic method, and execution steps of the Max-Max heuristic method comprise:

[0021] a pesticide spraying time sorting step, pesticide spraying times of all tasks are calculated, and all tasks are sorted in descending order of the pesticide spraying times, the task data comprises a pesticide demand of a task point, and the pesticide spraying times are determined according to the pesticide demand of the task point;

[0022] a pesticide loading amount sorting step, all pesticide spraying robots are sorted in descending order of pesticide loading amounts, and the pesticide spraying robot feature data comprises the pesticide loading amounts;

[0023] a task allocation step, each task is allocated to a corresponding optimal insertion position based on the pesticide spraying time window and the comprehensive cost value until all task allocation ends.

[0024] As a preferred technical solution, the optimal insertion position selection process comprises: comparing the delay penalty cost of assigning the current task to all insertable positions of each pesticide spraying robot, and selecting the position with the minimum delay penalty cost.

[0025] As a preferred technical solution, the current destruction intensity is generated based on the destruction intensity in the previous iteration round, and the current destruction intensity expression is:

[0026]

[0027] In the formula, Ruindisplays the current destruction intensity; Ruin ind indicates the initial value of the destruction intensity between the lower bound of the destruction intensity and the upper bound of the destruction intensity; ind ∈ {1, 2, …, Psize}, Psize is the population size; e N(0，1) is a standard normal distribution.

[0028] As a preferred technical solution, the destruction operator comprises a robot maximum delay penalty cost destruction operator, a robot maximum comprehensive cost destruction operator, a correlation destruction operator and a random destruction operator.

[0029] As a preferred technical solution, the destruction operation of the robot maximum delay penalty cost destruction operator comprises: selecting a pesticide spraying robot with the maximum delay penalty cost, randomly destroying tasks in the task list of the pesticide spraying robot, and randomly selecting tasks from the task lists of other pesticide spraying robots, wherein, Ruindisplays the current destruction intensity.

[0030] As a preferred technical solution, the destruction operation of the robot maximum comprehensive cost destruction operator comprises: selecting a pesticide spraying robot with the maximum comprehensive cost value, randomly destroying tasks in the task list of the pesticide spraying robot, and randomly selecting tasks from the task lists of other pesticide spraying robots, wherein, Ruindisplays the current destruction intensity.

[0031] As a preferred technical solution, the destruction operation of the correlation destruction operator comprises:

[0032] A first task destruction step, randomly selecting a current first task from all tasks, and moving the current first task to a destroyed task set;

[0033] A relevancy ranking step, traversing all second tasks, calculating the relevancy of the current first task with each second task, and ranking all relevancies from small to large, the second task is selected after the current first task The remaining tasks;

[0034] A second task destruction step, selecting the second task corresponding to the minimum relevancy, moving to the destroyed task set, and returning to the first task destruction step.

[0035] Compared with the prior art, the present application has the following beneficial effects:

[0036] 1、The present application obtains the feature data and task data of the pesticide spraying robot, combines heuristic method and adaptive destruction mechanism, and proposes an iterative greedy algorithm that meets the comprehensive requirements of global optimization, calculation time, solution quality, convergence speed, etc., which can automatically calculate the best solution of the task allocation and scheduling of the heterogeneous pesticide spraying robot, and when generating the solution, the pesticide spraying time constraint in the actual agricultural operation process and the performance constraint of the agricultural machinery center heterogeneous pesticide spraying robot are considered., can efficiently complete the allocation and scheduling of the pesticide spraying task in the appropriate pesticide spraying time interval, taking into account the different performance of heterogeneous robots, effectively improving the rationality of the heterogeneous robot pesticide spraying task planning, the difference in pesticide spraying effect, and reducing the cost of task allocation and scheduling;

[0037] 2、The present application uses an efficient heuristic method to generate an initial solution, which considers the performance of the heterogeneous robot and the pesticide spraying time window when allocating tasks, selects the best insertion position of the task, and can overcome the low quality of the initial solution generated by the traditional algorithm randomly;

[0038] 3、The present application considers the comprehensive cost value, which is helpful to solve the cost optimization problem in the pesticide spraying process of the heterogeneous pesticide spraying robot, especially to solve the comprehensive cost problem of multiple pesticide spraying robots, the comprehensive cost includes the delay penalty cost of completing the pesticide spraying task too late, the pesticide spraying robot driving cost and the pesticide spraying robot dispatching cost, thereby significantly improving the efficiency of the smart farm;

[0039] 4、The present application uses an adaptive destruction strength, and automatically matches a destruction operator according to the evolution state, which avoids the problem that a single destruction strength cannot adapt to all situations, and can further improve the quality and adaptability of the solution of the heterogeneous robot pesticide spraying task allocation and scheduling;

[0040] 5、The present application considers the pesticide spraying time, which is consistent with the limitation of pesticide spraying time in the actual agricultural environment, can reduce the comprehensive cost of solving the heterogeneous pesticide spraying robot task allocation and scheduling problem, thereby significantly improving the utilization rate of pesticides and the production efficiency of the smart farm, and reducing the comprehensive cost of the smart farm. BRIEF DESCRIPTION OF DRAWINGS

[0041] Figure 1 This is a schematic flowchart of the method of the present invention;

[0042] Figure 2 This is a table showing the ANOVA analysis results for all data in this embodiment of the invention.

[0043] Figure 3 This is a scatter plot showing the average of the comparison algorithms in this embodiment of the invention;

[0044] Figure 4 This is an interaction diagram showing the comparison algorithm and the number of spraying robots in an embodiment of the present invention;

[0045] Figure 5 This is an interaction diagram comparing the algorithm and task size in an embodiment of the present invention. Detailed Implementation

[0046] The present invention will now be described in detail with reference to the accompanying drawings and specific embodiments. These embodiments are based on the technical solution of the present invention and provide detailed implementation methods and specific operating procedures. However, the scope of protection of the present invention is not limited to the following embodiments.

[0047] Example

[0048] like Figure 1 As shown in the figure, this embodiment provides a method for task allocation and scheduling of heterogeneous spraying robots. The method includes a data acquisition step, a population initialization step, an adaptive destruction step, a solution reconstruction step, a local search optimization step, and an optimal solution acquisition step.

[0049] Step 1, Data Acquisition. This step includes acquiring characteristic data and task data of the spraying robots, as well as parameter initialization. The characteristic data of the spraying robots includes the number of robots, their starting position, moving speed, work efficiency, and pesticide load, etc. The task data includes the spraying time window, task size, task location, and pesticide requirement at each task point, etc. The damage intensity is the number of tasks removed. The parameter initialization process includes setting the population size Psize and the lower bound of the damage intensity Ruin. min Upper limit of destructive strength Ruin max The algorithm's stopping time T is defined as follows: the number of individuals in the population corresponds to the number of feasible solutions; the lower bound of the destruction strength represents the lower bound of the number of tasks destroyed in the destruction phase of the iterative greedy algorithm, while the upper bound of the destruction strength represents the upper bound of the number of tasks destroyed in the destruction phase; the stopping time T = 100 * m (where m is the task size), and the task size can also be obtained from the task data.

[0050] Step 2, Population Initialization. Encode the data using a task sequence-based representation method. Generate 3 individuals using three heuristic methods based on problem knowledge. The remaining Psize-3 individuals are generated randomly. This forms an initial population Π = {π1, π2, ..., π} containing Psize individuals. Psize}, select the individual with the lowest overall cost value from the initial population Π, and update the optimal solution.

[0051] Specifically, the comprehensive cost value TC ind The calculation method is as follows:

[0052] TC ind =TD ind +TR ind +C robot (1)

[0053] In the formula, TD ind TR ind C robot These are the costs of delays for the spraying robot (penalty costs for completing the task too late), the operating costs of the spraying robot, and the dispatch costs of the spraying robot.

[0054] Specifically, the encoding steps based on the task sequence representation method are as follows: An individual represents a complete solution and is composed of a two-dimensional vector. The first dimension of the two-dimensional vector consists of n one-dimensional vectors, where n represents the number of available spraying robots. One robot corresponds to one one-dimensional vector. Each one-dimensional vector contains multiple tasks waiting to be sprayed. The order of the tasks is consistent with the scheduling order of the spraying robots. Based on the above encoding rules, an individual can be represented as π = {π1, π2, ..., π}. n}, where the task sequence of the k-th spraying robot is k∈{1, 2, ..., n}, n k This represents the total number of tasks that the k-th spraying robot will perform.

[0055] Specifically, three problem-knowledge-based heuristics are designed to generate three individuals, aiming to provide a high-quality initial solution for the iterative greedy algorithm. These three problem-knowledge-based heuristics are the fastest arrival heuristic, the fastest completion heuristic, and the Max-Max heuristic.

[0056] (1) Fastest arrival heuristic method

[0057] The fastest arrival heuristic method prioritizes the tasks closest to the pesticide spraying robot warehouse, aiming to greedily solve the closest tasks, focusing on optimizing the penalty cost of the pesticide spraying robot due to completing tasks too late in the pesticide spraying time window of the pesticide spraying robot task allocation and scheduling problem. The specific steps are as follows:

[0058] Arrival time sorting step, calculate the distance of all task points from the pesticide spraying robot starting point, and calculate the arrival time of the pesticide spraying robot, sort all tasks in ascending order of arrival time, and the arrival time is calculated using the aforementioned distance and the moving speed of the pesticide spraying robot;

[0059] Task allocation step, extract the sorted tasks one by one, and assign them to the corresponding optimal insertion position based on the pesticide spraying time window until all tasks are successfully assigned, finally generating an individual (a complete solution). Among them, the optimal insertion position based on the pesticide spraying time window is the position with the minimum penalty cost of delaying the completion of tasks due to the selected pesticide spraying robot.

[0060] (2) Fastest completion heuristic method

[0061] The fastest completion heuristic method prioritizes the pesticide demand of tasks to be executed, aiming to greedily solve more tasks once the pesticide spraying robot is dispatched, focusing on optimizing the penalty cost of the pesticide spraying robot due to completing tasks too late in the pesticide spraying time window of the pesticide spraying robot task allocation and scheduling problem. The specific steps are as follows:

[0062] Pesticide spraying time sorting step, calculate the pesticide spraying time of all tasks, generally, the smaller the pesticide spraying area, the smaller the pesticide demand and the shorter the working time, sort all tasks in ascending order of pesticide spraying time;

[0063] Task allocation step, extract the sorted tasks one by one, and assign them to the corresponding optimal insertion position based on the pesticide spraying time window until all tasks are successfully assigned to specific pesticide spraying robots, finally generating an individual (a complete solution). Among them, the optimal insertion position based on the pesticide spraying time window is the position with the minimum penalty cost of delaying the completion of tasks due to the selected pesticide spraying robot.

[0064] (3) Max-Max heuristic method

[0065] In the task allocation and scheduling problem of heterogeneous pesticide spraying robots, there are multiple robots with different working efficiency and working capacity. In order to maximize the use of multiple heterogeneous pesticide spraying robots, the Max-Max heuristic method preferentially allocates the task with the largest pesticide demand to the pesticide spraying robot with the largest execution capacity. For the pesticide spraying time window in the task allocation and scheduling problem of pesticide spraying robots, the focus is on optimizing the penalty cost of the pesticide spraying robot due to the task completed too late. The specific steps are as follows:

[0066] Pesticide spraying time sorting step, calculate the pesticide spraying time of all tasks, in general, the smaller the pesticide spraying area, the smaller the pesticide demand and the shorter the working time, sort all tasks according to the pesticide spraying time from large to small;

[0067] Pesticide loading capacity sorting step, sort all pesticide spraying robots according to the pesticide loading capacity from large to small;

[0068] Task allocation step, preferentially allocate the task with the largest pesticide spraying time to the pesticide spraying robot with the largest pesticide loading capacity, allocate the task with the second largest pesticide spraying time to the pesticide spraying robot with the second largest pesticide loading capacity, and allocate the task with the nth largest pesticide spraying time to the pesticide spraying robot with the nth largest pesticide loading capacity.

[0069] Next, extract the remaining tasks n+j (j≥1) in turn, repeat the above process, and allocate them to the pesticide spraying robot R k , and based on the pesticide spraying time window, allocate the task to the best insertion position of the pesticide spraying robot R k , until all tasks are allocated, and finally generate an individual (a complete solution). Among them, the best insertion position based on the pesticide spraying time window is the position with the smallest delay penalty cost due to the task completed too late after the task is allocated to all possible positions of a pesticide spraying robot.

[0070] In addition, the remaining Psize-3 individuals are generated by a random method. The specific steps of the random method are as follows:

[0071] Randomly select n tasks, allocate n tasks to n pesticide spraying robots in a random way; the remaining tasks are randomly extracted, one task is extracted at a time, and it is allocated to the best insertion position, and finally an individual is generated. The best insertion position is the position with the smallest delay penalty cost due to the task completed too late after the task is allocated to all possible positions of each pesticide spraying robot.

[0072] Step 3, adaptive destruction step. According to the current evolutionary state, match the destruction strength initial value and the destruction operator for each solution, and perform the destruction operation according to the destruction operator and the current destruction strength, that is, destroy (perturb) the individual through the destruction mechanism, so as to obtain the current remaining task set and the destruction task set, and the destruction operator is related to the feature data or task data of different spraying robots.

[0073] For automatic matching of destruction strength: the destruction strength parameter of the iterative greedy algorithm has a great influence on performance, and a single fixed destruction strength parameter is not universal, and a single fixed destruction strength parameter may not be effective for some task sizes, spraying robot quantities, and work efficiency. Therefore, in step 3, the iterative greedy algorithm uses an adaptive destruction strength parameter, and the adaptive destruction stage first needs to determine the destruction strength of the individual in each iteration process, and the specific steps are as follows:

[0074] Destruction strength initial value matching step: each individual corresponds to an initial destruction strength before evolution starts:

[0075] Ruin ind ∈[Ruin min ,Ruin max ],ind∈{1,2,…Psize} (2)

[0076] In the formula, Ruin ind represents the destruction strength initial value, between the lower bound of the destruction strength and the upper bound of the destruction strength; ind∈{1,2,…Psize}, Psize is the population size;

[0077] Current destruction strength generation step: at each step of the evolution process, a new current destruction strength will be generated based on the destruction strength of the last iteration process, and the expression is:

[0078]

[0079] In the formula, e N(0，1) is a standard normal distribution. The above destruction strength refers to the number of tasks removed (destroyed) from the current solution.

[0080] For automatic matching of destruction operator: in step 3, the destruction mechanism will select a suitable destruction operator according to the state of the individual in the evolution process, and the four destruction operators based on the characteristics of the heterogeneous spraying robot task allocation and scheduling problem are the robot maximum delay penalty cost destruction operator, the robot maximum comprehensive cost destruction operator, the correlation destruction operator, and the random destruction operator.

[0081] (1) Robot maximum delay penalty cost destruction operator

[0082] Robot maximum delay penalty cost destruction operator focuses on the delay penalty cost affecting the comprehensive cost value. Serious delay not only makes the comprehensive cost value rise, but also seriously affects the drug efficacy. Therefore, paying attention to the pesticide spraying robot with the maximum delay cost will help obtain a better allocation and scheduling scheme. The specific steps are as follows:

[0083] Select the pesticide spraying robot Robot mt with the maximum delay penalty cost mt Randomly destroy tasks from Robot Select the remaining tasks from the task list of other pesticide spraying robots, and obtain the remaining task set and the destroyed task set

[0084] (2) Robot maximum comprehensive cost destruction operator

[0085] The robot maximum comprehensive cost destruction operator focuses on the pesticide spraying robot with the maximum comprehensive cost value, aiming to more specifically destroy tasks and optimize the robot task sequence, and generate Π R and Π D . The specific steps are as follows:

[0086] Select the pesticide spraying robot Robot mz with the maximum comprehensive cost mz Randomly destroy tasks from Robot Select the remaining tasks from the task list of other pesticide spraying robots, and obtain the remaining task set and the destroyed task set

[0087] (3) Related destruction operator

[0088] According to the characteristics of the drug demand of the task point and the location distribution, the related destruction operator aims to destroy a group of tasks with high correlation. The specific steps are as follows:

[0089] First task destruction step, the first task in this embodiment is represented by J i . Randomly select a task J i and move it to the destroyed task set

[0090] Correlation sorting step, the second task in this embodiment is represented by J j . Traverse all second tasks, the second task is the remaining task after J i is selected, calculate the correlation RD i between the task J j and J ijThen the task relevances are sorted in ascending order, RD ij The smaller the value, the greater the relevance between the two tasks.

[0091] The second task destruction step selects the task J ij with the smallest RD j value and moves it to the destroyed task set The next task is iteratively selected until all tasks are destroyed.

[0092] The calculation formula of the relevance RD ij is

[0093] RD ij = δ + 3 * |D i -D j | + 2 * C ij (4)

[0094] If two tasks are assigned to the same spraying robot, then δ = 1, otherwise δ = -1; D i , D j represent the spraying area of tasks J i , J j respectively; and C ij represents the distance from task J i to task J j .

[0095] (4) Random destruction operator

[0096] The specific steps of the random destruction operator are as follows:

[0097] Randomly destroy tasks from the task list of all spraying robots to obtain the remaining task set and the destroyed task set

[0098] The adaptive destruction mechanism used in this embodiment will select the above destruction operators to perform destruction operations according to the evolutionary state of the individual (solution). The specific steps are as follows:

[0099] If an individual is improved in one iteration evolution, set isImp as tme, otherwise set isImp as false; in the next iteration, if isImp = true, then randomly select one of the robot maximum comprehensive cost destruction operator and the robot maximum delay penalty cost destruction operator to perform further destruction operation; otherwise, randomly select one of the above four destruction operators to perform destruction operation.

[0100] Step 4, solution reconstruction step, sequentially assign all tasks in the current destroyed task set to the position with the minimum integrated cost value in the remaining task set to obtain the current solution. That is, reconstruct the destroyed solution (solution), sequentially assign tasks in the destroyed task set Π D to Π R , and finally generate a complete new individual (new solution).

[0101] Specifically, tasks are extracted from the destroyed task set one by one, and attempts are made to assign them to all possible positions in the remaining task set , and the integrated cost is calculated. Assign the task to the position with the minimum integrated cost until all tasks in are reassigned to the optimal position. The optimal position refers to the position with the minimum integrated cost.

[0102] Step 5, local search optimization step, through local search, reassign all tasks of each spraying robot one by one to obtain a solution with the minimum integrated cost. The local search method can further enhance the local exploitation ability of the algorithm and improve the quality of the individual, that is, for each spraying robot, perform deep optimization, and the specific steps include:

[0103] Step 51, extract the first task assigned to the kth spraying robot;

[0104] Step 52, attempt to assign the task to other positions in the task list of the spraying robot, and select the best assignment position from all positions;

[0105] Step 53, insert the task into the optimal position and obtain a new solution (new solution). If the integrated cost of the new solution is better than that of the old solution, replace the old solution with the new solution, and reextract the first task of the robot. Otherwise, continue to extract the next task for reassignment;

[0106] Step 54, repeat steps 51-53 until all tasks in the task list of the robot are reassigned.

[0107] Step 6, best solution acquisition step. Determine whether the stop time has been reached. If not, return to step 3. If yes, output the current solution as the best solution to obtain the best heterogeneous spraying robot task assignment and scheduling scheme.

[0108] It should be noted that in steps 1-6, all spraying robots start from the warehouse, and these spraying robots may have different drug carrying capacities, work efficiencies, travel speeds, etc. due to technical differences, i.e. belong to heterogeneous spraying robots. During the execution of the task, if the remaining amount of carried pesticide is insufficient to complete the next task, the spraying robot returns to the warehouse to add pesticide.

[0109] In this embodiment, the method provided is applied to a spraying robot warehouse, multiple spraying robots of different models (heterogeneous), and the goal is to determine a best solution considering the spraying time window (9:00-11:00 in the morning, 16:00-19:00 in the afternoon) to minimize the total cost. The population size Psize is 30 (the population contains 30 individuals); the lower bound of the destruction strength Ruin min = 1 (the lower bound of the number of tasks destroyed (removed) from the individual each time); the upper bound of the destruction strength Ruin max = 5 (the lower bound of the number of tasks destroyed (removed) from the individual each time); the stopping time T of the algorithm is 100*m milliseconds (m is the task size) as an example, which is described in detail and verified by experiment.

[0110] The standard test example configuration used in the simulation experiment is as follows: the number of spraying robots is n = {3, 5, 7, 9, 11, 13,}, the task size is m = {30, 40, 60, 80}, the task point pesticide demand is D = {[5, 12], [10, 17]}, the work efficiency of the heterogeneous spraying robot is prod-rate = {4.8, 7.8, 33, 61.2}, the moving speed of the heterogeneous spraying robot is speed = {60, 72, 90, 120}, and the single load capacity of the heterogeneous spraying robot is Capa = {6, 12, 25, 35}. The iterative greedy algorithm (DIG) proposed in this embodiment is compared with several advanced evolutionary algorithms to verify the effectiveness of the algorithm proposed in this embodiment. These several comparison algorithms are high-performance evolutionary algorithms proposed in recent years, which are genetic algorithm (GA), discrete artificial bee colony algorithm (DABC1), and artificial bee colony algorithm (ABC).

[0111] A specific example is taken as an example to show the solutions obtained by the DIG algorithm proposed in this embodiment and several advanced evolutionary algorithms. The example includes 5 heterogeneous spraying robots (n = 5) and 60 tasks (m = 60). Table 1 shows the performance parameters of the 5 robots, especially the work efficiency, capacity, and travel speed. Table 2 is the solution result of all algorithms, where the "solution" lists the tasks allocated to each spraying robot and the task execution order of each robot considering the spraying time window constraint and the different work capacity constraints of the heterogeneous robots.

[0112] Table 1 Heterogeneous spraying robot performance parameters (n = 5)

[0113]

[0114] Table 2 Solution example (5 heterogeneous robots (n = 5), 60 tasks (m = 60))

[0115]

[0116]

[0117] As can be seen from Table 2, the method proposed in the embodiment obtains a better quality solution (obtains the lowest comprehensive cost 1460.1), and is obviously superior to the solutions obtained by the comparative algorithms (ABC (1704.7), DABC1 (1498.65), GA (2044.38)). The method proposed in the embodiment can fully consider the working capacity of different spraying robots (the No. 1 robot has high speed, high working efficiency and large capacity, and is reasonably allocated more tasks), and optimizes the comprehensive cost to the maximum extent.

[0118] In order to eliminate the contingency of the experiment and reduce the experimental error, so that the calculation result has more statistical significance, each example in the experimental link is repeated 5 times to generate statistical results. The relative percentage increase (RPI) is used as the performance evaluation index, and the calculation formula of RPI is

[0119]

[0120] Among them, TC represents the comprehensive cost value obtained by a specific algorithm when solving an example, and TC best represents the minimum comprehensive cost value obtained by solving the same example among the four algorithms. Obviously, the smaller the RPI value, the smaller the comprehensive cost found by the algorithm, and the better the performance.

[0121] Figure 2 is the multi-factor variance analysis (ANOVA) of all data, Figure 3 , Figure 4 and Figure 5 is the intuitive representation of the optimization results of all algorithms. Specifically, from the multi-factor variance analysis table of Figure 2 , the significance (p-value) is equal to 0.0, and since the significance statistical value is less than 0.05 (i.e. 95% confidence interval), all algorithm performances have significant differences. In order to intuitively observe the differences between algorithms, Figures 3 to 5 shows the average scatter plot (95% HSD confidence interval) of different algorithms and the algorithm interaction diagram. From Figure 3It can be seen that, among all the algorithms, DIG obtains the best RPI value (the smallest RPI value), indicating that DIG has the best performance and is obviously superior to the comparative algorithms. Figure 4 and Figure 5 It can be seen from the interaction diagram of and that, for different task sizes and different numbers of spraying robots, DIG obtains better results than the comparative algorithms, and the performance of DIG is more stable (the interaction curve fluctuates less and is more stable). In summary, the simulation experiments on 48 examples verify the superiority and effectiveness of the method proposed in the embodiment.

[0122] The preferred embodiments of the application are described in detail above. It should be understood that those of ordinary skill in the art can make many modifications and changes without creative labor based on the concept of the application. Therefore, any technical solutions obtained by logical analysis, reasoning or limited experiments based on the existing technology according to the concept of the application shall be within the protection scope defined by the claims.

Claims

1. A heterogeneous pesticide spraying robot task allocation and scheduling method, characterized in that, The method comprises the following steps: a data acquisition step of acquiring feature data of the pesticide spraying robots and task data, and initializing a population size, a lower bound of damage intensity, an upper bound of damage intensity, and a stop time, the task data including a pesticide spraying time window, and the damage intensity being a number of removed tasks; a population initialization step of generating multiple initial solutions by using a heuristic method and a random method respectively, and calculating corresponding comprehensive cost values, forming an initial population, and selecting an initial solution with the minimum comprehensive cost value from the initial population, the heuristic method including assigning tasks based on the feature data of the pesticide spraying robots and the task data, and selecting optimal insertion positions of the tasks based on the pesticide spraying time window and the comprehensive cost value, the comprehensive cost value being a sum of a pesticide spraying robot delay penalty cost, a travel cost, and a dispatching cost; an adaptive damage step of matching a damage operator with an initial damage intensity value according to a current evolution state, performing a damage operation according to the damage operator and the current damage intensity, and acquiring a current remaining task set and a damage task set, the damage operator being related to the feature data or the task data of different pesticide spraying robots; a solution reconstruction step of sequentially assigning all tasks in the current damage task set to positions with the minimum comprehensive cost value in the remaining task set to obtain a current solution; a local search optimization step of sequentially reassigning all tasks of each pesticide spraying robot by local search to obtain a solution with the minimum comprehensive cost value; a best solution acquisition step of judging whether the stop time is reached, returning to the adaptive damage step if the stop time is not reached, and outputting the current solution if the stop time is reached to obtain a best heterogeneous pesticide spraying robot task assignment and scheduling scheme.

2. The heterogeneous pesticide spraying robot task allocation and scheduling method according to claim 1, characterized in that, The heuristic method includes a fastest arrival heuristic method, and the execution steps of the fastest arrival heuristic method include: an arrival time sorting step of calculating distances of positions of all task points from a starting position of a pesticide spraying robot and arrival times of the pesticide spraying robot, sorting all tasks in ascending order of the arrival times, the feature data of the pesticide spraying robot including the starting position and a moving speed of the pesticide spraying robot, and the task data including positions of the task points, the arrival times being calculated based on the distances and the moving speed; a task assignment step of sequentially extracting each task sorted in ascending order of the arrival times, assigning each task to a corresponding optimal insertion position based on the pesticide spraying time window and the comprehensive cost value, and ending the assignment of all tasks.

3. The method of claim 1, wherein, The heuristic method includes a fastest completion heuristic method, and the execution steps of the fastest completion heuristic method include: a pesticide spraying time sorting step of calculating pesticide spraying times of all tasks, sorting all tasks in ascending order of the pesticide spraying times, the task data including pesticide demand amounts of the task points, and the pesticide spraying times being determined based on the pesticide demand amounts; a task assignment step of sequentially extracting each task sorted in ascending order of the pesticide spraying times, assigning each task to a corresponding optimal insertion position based on the pesticide spraying time window and the comprehensive cost value, and ending the assignment of all tasks.

4. The method of claim 1, wherein, The heuristic method includes a Max-Max heuristic method, and execution steps of the Max-Max heuristic method include: a pesticide spraying time sequencing step of calculating pesticide spraying times of all tasks, and sequencing all tasks in descending order of the pesticide spraying times, wherein the task data includes a task point pesticide demand, and the pesticide spraying time is determined according to the task point pesticide demand; a pesticide loading amount sequencing step of sequencing all pesticide spraying robots in descending order of pesticide loading amounts, wherein the pesticide spraying robot feature data includes the pesticide loading amount; a task allocation step of extracting each task sequenced in descending order of the pesticide spraying times one by one, and allocating each task to each pesticide spraying robot sequenced in descending order of the pesticide spraying times one by one, wherein the task is allocated to an optimal insertion position of the corresponding pesticide spraying robot based on the pesticide spraying time window and the comprehensive cost value.

5. The method of claim 1-4, wherein, The optimal insertion position selection process includes comparing postponement penalty costs of all insertable positions of each pesticide spraying robot to which the current task is allocated, and selecting a position with the minimum postponement penalty cost.

6. The method of claim 1, wherein, The current damage strength is generated based on the damage strength in the previous iteration round, and the current damage strength expression is: wherein Ruin represents the current ruin strength; Ruin ind ind represents the ruin strength initial value, between the lower bound of the ruin strength and the upper bound of the ruin strength; ind ∈ {1, 2, … Psize}, Psize is the population size; e N(0,1) is a standard normal distribution.

7. The method of claim 1, wherein, The damage operator includes a robot maximum postponement penalty cost damage operator, a robot maximum comprehensive cost damage operator, a correlation damage operator and a random damage operator.

8. The method of claim 7, wherein, The destruction operation of the robot maximum delay penalty cost destruction operator includes selecting a pesticide spraying robot with the maximum delay penalty cost, and randomly destroying one task in the task list of the pesticide spraying robot The remaining tasks are randomly selected from the task lists of other pesticide spraying robots, wherein represents the current destruction intensity.

9. The method of claim 7, wherein the method further comprises: The destroying operation of the robot maximum comprehensive cost destroying operator includes selecting a pesticide spraying robot with the maximum comprehensive cost value, and randomly destroying one task in the task list of the pesticide spraying robot The remaining tasks are randomly selected from the task lists of other pesticide spraying robots, wherein The current destroying strength is represented by ​ 10. The method of claim 7, wherein, The damage operation of the correlation damage operator includes: a first task damage step of randomly selecting a current first task from all tasks, and moving the current first task to a damage task set; a correlation degree sequencing step of traversing all second tasks, calculating correlation degrees of the current first task and each second task, and sequencing all correlation degrees in ascending order, wherein the second task is a remaining task after the current first task is selected; a second task damage step of selecting a second task corresponding to the minimum correlation degree, moving the second task to the damage task set, and returning to the first task damage step.

Images