Multi-equipment task intelligent decomposition and scheduling method based on target point clustering and improved genetic algorithm

Abstract

The invention provides a multi-equipment task intelligent decomposition and scheduling method based on target point clustering and an improved genetic algorithm. The method comprises the following steps: S1, carrying out element modeling on a multi-equipment task; the elements comprise equipment, tasks, target points and time; s2, decomposing a multi-equipment task based on target point clustering; s3, constructing a multi-equipment task scheduling model; and S4, solving the multi-equipment task scheduling model based on an improved multi-target genetic algorithm. According to the method, target point clustering is combined, interval numbers are used for representing uncertain parameters, and an index task fine-grained multi-target planning model is established by taking the equipment capacity as far as possible meeting task requirements and the task completion time as short as possible as optimization indexes; and then a multi-objective genetic optimization algorithm is improved by using the congestion degree of a solution space and the dominating relation of a non-inferior solution, the solving efficiency is improved, and the purpose of quickly and adaptively determining an equipment task scheduling scheme is achieved.

Description

Technical Field

[0001] The invention relates to the field of equipment task allocation, in particular to a multi-equipment task intelligent decomposition and scheduling method based on target point clustering and improved genetic algorithm. Background Art

[0002] Equipment task decomposition and scheduling is the process of decomposing the overall task that needs to be performed by the equipment into specific tasks, and then allocating and optimizing the tasks in time. Due to the diversity of equipment types and task types, the problem of intelligent decomposition and scheduling of equipment tasks is a difficult problem in the field of equipment task planning. In the prior art, there are task decomposition technology based on hierarchical task networks and planning model solving technology based on NSGA-II genetic algorithm.

[0003] (1) Task decomposition technology based on hierarchical task network

[0004] Hierarchical Task Network (HTN) is a knowledge-based planning technology and one of the most widely used rule-based methods. It expresses the hierarchical relationship between tasks by abstracting and formalizing the tasks in a hierarchical manner, and is therefore more suitable for the hierarchical decomposition of complex tasks.

[0005] The basic principle of the hierarchical task network is to decompose the tasks layer by layer, starting with the compound task, first decomposing it into simpler subtasks, and then continuing to decompose the subtasks one by one until all tasks are decomposed into directly executable atomic tasks. The whole decomposition process is completed, and finally a hierarchical tree-like task network composed of compound tasks, subtasks, and atomic tasks is formed, thereby realizing the detailed decomposition of tasks and converting complex tasks into executable simple tasks.

[0006] In a hierarchical task network, a task is defined as the work that needs to be completed to achieve a specific goal, and tasks are divided into three types: complex tasks, subtasks, and atomic tasks. Complex tasks are at the top level of the task network and are the objects of task decomposition. They are complex tasks that cannot be executed directly and are usually composed of several atomic tasks according to certain logical relationships and structures. Subtasks are located in the middle layer of the task network and are intermediate tasks formed in the process of task decomposition. Atomic tasks are at the bottom layer of the task network and are the final end tasks formed by task decomposition. They do not need to be decomposed further and usually contain only one action and can be executed directly.

[0007] In the HTN-based task decomposition framework, composite tasks are the objects of task decomposition and define the starting point of task decomposition; atomic tasks are the final results of task decomposition and determine the end point of task decomposition. For different task decomposition scenarios, the starting point and end point of task decomposition are often different. Different task decomposition starting points and end points correspond to different levels of task decomposition complexity. Therefore, when using HTN for task decomposition, we must first clarify the level of task decomposition, and based on this, analyze the composite tasks and atomic tasks in specific task decomposition scenarios, and then perform task decomposition in a targeted manner.

[0008] The basic idea of ​​equipment task hierarchical decomposition is to comprehensively consider factors such as the structural characteristics of the task, environmental constraints, decomposition methods, and the relationship between subtasks, and to break down the overall task into a number of specific, clearly hierarchical, independent (the degree of coordination between subtasks at the same level is as low as possible), and easy-to-execute subtasks. The subtasks at the upper level play a dominant role on all or some of the subtasks at the lower level (high coordination between tasks at the upper and lower levels). Finally, a hierarchical tree-like hierarchical structure is obtained through comprehensive arrangement.

[0009] (2) Planning model solving technology based on NSGA-II genetic algorithm

[0010] It is often difficult to obtain a solution that achieves optimality in all objectives for multi-objective planning problems. Therefore, the optimality of such problems often refers to the existence of a set of "non-inferior solutions" that cannot be compared with each other. This is the Pareto optimal solution.

[0011] In multi-objective optimization problems, Pareto optimal solution is the most commonly used optimization concept. Taking the minimization problem as an example, its definition is as follows: Let F(x) = (F1(x), F2(x), …, F m (x)) T is the n objective functions of the multi-objective optimization problem. Assume that x1 and x2 are two feasible solutions in the solution space if and only if F i (x1)≤F i (x2), i = 1, 2, ..., n, x1 is said to be non-dominated by x2, or x1 dominates x2. If there is no dominating x in the solution space * The solution of x is * is the Pareto optimal solution. Therefore, the "optimal solution" that satisfies this optimality is often not a single solution, but a set of non-inferior solutions that meet the optimality conditions.

[0012] The non-dominated sorting genetic algorithm is a genetic algorithm based on the concept of Pareto optimality. Among all the multi-objective genetic algorithms based on Pareto optimal solutions, NSGA is the most direct method to embody Goldberg's idea.

[0013] The main difference from the simple genetic algorithm is that the NSGA algorithm stratifies the individuals in the population according to the dominance relationship before the selection operator is executed, while the selection, crossover and mutation operators of the NSGA algorithm are not much different from the simple genetic algorithm. NSGA retains excellent individuals through the non-dominated sorting algorithm and maintains the diversity of the population by using the fitness sharing function method. NSGA-Ⅱ is an improvement on NSGA, which reduces the complexity of the algorithm and increases the probability of algorithm convergence.

[0014] The basic steps of the NSGA-II algorithm are: First, randomly generate an initial parent population P of size N t , through crossover mutation to generate offspring population Q t ; Combine the two parent populations and the offspring population to form a new population R of size 2N t , and for R t Perform fast non-dominated sorting and calculate the crowding degree of individuals in each non-dominated layer. Select appropriate individuals to form a new parent population P according to the non-dominated relationship and the crowding degree of the individuals. t+1 ; Then repeat the above steps until the conditions for the program to end are met. Summary of the invention

[0015] The purpose of the present invention is to provide a multi-equipment task intelligent decomposition and scheduling method based on target point clustering and improved genetic algorithm, which can automatically realize equipment task decomposition and improve the timeliness of equipment task scheduling.

[0016] The present invention is implemented as follows: a multi-equipment task intelligent decomposition and scheduling method based on target point clustering and improved genetic algorithm, comprising the following steps:

[0017] Step S1, modeling the elements of a multi-equipment task; the elements include equipment, tasks, target points and time;

[0018] Step S2, decomposing the multi-equipment tasks based on target point clustering;

[0019] Step S3, constructing a multi-equipment task scheduling model;

[0020] Step S4, solving the multi-equipment task scheduling model based on the improved multi-objective genetic algorithm; Step S4 is specifically as follows:

[0021] Step S4-1: adopting a chromosome decoding method that satisfies the model constraints to ensure that the reachable space of the algorithm corresponds to the feasible domain of the model;

[0022] Step S4-2: The dominance relationship of the number of intervals based on the interval possibility is used for comparison and ranking of model targets;

[0023] Step S4-3: Calculate the crowding distance of the solution using the interval distance formula, and use cyclic crowding sorting to eliminate redundant solutions;

[0024] Step S4-4: Design a heuristic crossover operator to guide population evolution, improve the convergence speed of the algorithm, and ultimately obtain a solution to multi-equipment task scheduling.

[0025] In the above scheme, step S2 is specifically as follows:

[0026] Step S2-1: Determine the location information of all target points and the number of clusters N to be divided K , randomly select N from all target point positions K The positions are taken as the initial clustering centers;

[0027] Step S2-2: For all target points, calculate their differences from these N K The distance between the cluster centers is divided into clusters with the closest cluster center.

[0028] Step S2-3: In each newly divided cluster, calculate the average value of all positions in the cluster and use it as the new cluster center point;

[0029] Step S2-4: Repeat steps S2-2 and S2-3 until the clustering of the target point position no longer changes.

[0030] In the above scheme, step S3 is specifically as follows:

[0031] Step S3-1: Model the equipment capability requirements; record the equipment task set as T = {T1, T2, ···, T n}, n is the number of tasks in the set; for Its properties include: task coordinate location (XT i ,YT i ); Task duration TT i ;Task capability requirement vector (DT i1 ,DT i2 ,···,DT il );1≤k≤l;

[0032] Step S3-2: Model the equipment capability; the set of all equipment is denoted as A = {A1, A2, ···, A v}, v is the number of equipment; for Its properties include: the initial coordinate position of the equipment (XA j ,YA j ), equipment capability vector (OA j1 ,OA j2 ,...,OA jl), the equipment's movement speed VA j ; 1≤k≤l;

[0033] The task allocation scheme of the equipment group is represented by the matrix TAM=(tam ij ) n×v Indicates that if task T i Assigned to Equipment A j Processing, then tan i1 =1, otherwise tam ij =0;

[0034] Step S3-3: Modeling interval uncertainty parameters for equipment capability requirements and equipment capabilities; that is, providing intervals for each parameter by estimation method;

[0035] Step S3-4: Model the execution sequence of multiple equipment tasks; the timing constraint relationship between tasks in the task set T is represented by the matrix TCM=(tcm ij ) n×n Indicates; if task T i It is task T j The predecessor task, then tcm ij =1, otherwise tcm ij =0;

[0036] The task priority is calculated using the weighted length algorithm WL. The specific formula is as follows:

[0037]

[0038] Where WL(i) is the task T obtained by the WL algorithm. i The priority coefficient of task T j The priority coefficient, T j It is T i The direct successor task, OUT(i) is the task sequence diagram T i The set of tasks that are the direct successors of Indicates T i The maximum value of the task priority coefficient in the direct successor task set, ∑ j∈OUT(j) WL(j) represents T i The sum of the priority coefficients of all tasks in the direct successor task set;

[0039] Step S3-5: Model the multi-equipment task scheduling based on the task completion time and completion quality.

[0040] In the above scheme, step S3-5 is specifically as follows:

[0041] for The start time is recorded as BT i , the end time is recorded as ETi =BT i +TT i , the set of its predecessor tasks is denoted as ST Ti = {T k |tcm ki=1,1≤k≤n},in processing task T i When , the set of required capability types is denoted as R Ti ={k|DT ik >0, 1≤k≤u}, process T i The set of equipment is denoted as SA Ti = {A k |tam ik=1,1≤k≤u};

[0042] The objective function of the multi-equipment task scheduling model is:

[0043] min TM; max QM

[0044] TM is the time to complete all tasks, QM is the quality of completing all tasks;

[0045] for If it is processing T i If no other tasks have been processed before, then it reaches task T i The time at your location is TIME ki =((XA k -XT i ) 2 +(YA k -YT i ) 2 ) 1 / 2 / VA k ; If it is processing T i The last task processed before was T j , then it reaches T i The time at your location is TIME ki =((XT j -XT i ) 2 +(YT j -YT i ) 2 ) 1 / 2 / VA k +ET j ;

[0046] The multi-equipment task scheduling model is min f = (TM, 1-QM),

[0047]

[0048] Among them, tam ij is the decision variable of the model.

[0049] In the above scheme, step S4-1 is specifically as follows:

[0050] Chromosome X is encoded in binary code, consisting of an n-row and u-column 0-1 matrix CM=(cm ij ) n×u It means that each row of the matrix CM is a gene of the chromosome;

[0051] Step S4-1-1: Initialize i=0;

[0052] Step S4-1-2: Set i=i+1. If i<n+1, set k=0 and execute step S4-1-3. Otherwise, the decoding process ends.

[0053] Step S4-1-3: Set k=k+1. If k<l+1, execute step S4-1-4. Otherwise, set j=0 and execute step S4-1-5.

[0054] Step S4-1-4: If DT ik >0 and This means that when the task allocation scheme TAM = CM, the constraint QT i > 0 is not satisfied, and it is necessary to find the equipment A with the kth ability and the smallest serial number value in the equipment set A. q , and let cm iq =1, then return to step S4-1-3, otherwise directly return to step S4-1-3;

[0055] Step S4-1-5: Let j=j+1. If j<u+1, execute step S4-1-6, otherwise return to step S4-1-2;

[0056] Step S4-1-6: If cm ij =1, execute step S4-1-7, otherwise return to step S4-1-5;

[0057] Step S4-1-7: Let TAM = CM, and calculate c1 = QT based on TAM i Then let tam ij = 0, and calculate c2 = QT according to TAM i ; If c1=c2, let cm ij =0 and return to step S4-1-5, otherwise directly return to step S4-1-5;

[0058] The chromosome X=CM obtained after the decoding operation is a feasible solution of the model. At this time, CM is set as the task allocation scheme TAM.

[0059] In the above solution, in step S4-3, chromosomes entering the next-generation population are selected by sorting the crowding distances in a loop, that is: first, calculate the crowding distances of each alternative chromosome, then select the chromosome with the largest crowding distance into the next-generation population and delete it from the alternative chromosomes, and continuously repeat the above steps until the scale of the next-generation population reaches the upper limit;

[0060] When calculating the crowding distance of a chromosome, the interval distance formula used is:

[0061] d(A, B) = ((a L +a R -b L -b R ) 2 / 4 + ((a R -a L ) 2 +(b R -b L ) 2 ) / 12-F 2 / 6) 1 / 2

[0062] where: A = [a L , a R , B = [b L , b R ; F is the length of the interval A ∩ B, and when , F = 0; the dimensions of the two objective functions TM and 1 - QM are different, so the objective functions need to be normalized before calculating the crowding distance.

[0063] In the above solution, step S4-4 is specifically as follows:

[0064] Step S4-4-1: Select the chromosomes X1 = CM1 and X2 = CM2 participating in crossover, initialize the new chromosome X3 = CM3, where all elements of the matrix CM3 are 1, and set i = 1;

[0065] Step S4-4-2: If i < n + 1, then execute step S4-4-3, otherwise end and output X3 as the constructed new chromosome;

[0066] Step S4-4-3: Replace the i-th gene of X3 with the i-th genes of X1 and X2 respectively, and denote the formed new chromosomes as X 31 , X 32 ; if X 31 > X 32 , then let X3 = X 31 ; if X 31 < X 32 , then let X3 = X 32 ; if X31 ~X 32 , then from X 31 , X 32 Randomly select an update X3; let i=i+1 and return to step S4-4-2.

[0067] The multi-equipment task intelligent decomposition and scheduling method based on target point clustering and improved genetic algorithm proposed in this invention is superior to the traditional method in terms of equipment task decomposition accuracy, automation degree and timeliness of equipment task scheduling. Specifically, it is embodied in:

[0068] (1) The traditional task decomposition method (HTN) generally decomposes and refines specific tasks layer by layer. Considering the various types of equipment tasks and complex decomposition rules, this process usually needs to be completed manually. In order to achieve the purpose of intelligent decomposition, the present invention focuses on the core influencing factor of the target point, comprehensively considers the location information and other attributes of the target point, decomposes the equipment task based on the target point clustering algorithm, and obtains the decomposed subtask set.

[0069] (2) Equipment scheduling problems involve a variety of equipment types and complex task requirements. The model solution speed requirements of traditional equipment task scheduling methods are often difficult to meet the timeliness requirements. The present invention proposes an improved multi-objective genetic optimization algorithm to solve this problem. First, a fine-grained multi-objective planning model for indicator tasks is established with the equipment capability meeting the task requirements as much as possible and the task completion time being as short as possible as the optimization indicators; then, the multi-objective genetic optimization algorithm is improved by using the congestion of the solution space and the dominance relationship of non-inferior solutions, thereby improving the solution efficiency and achieving the purpose of quickly and adaptively determining the equipment task scheduling solution.

[0070] (3) In the equipment task scheduling work, the equipment capacity requirements and equipment transportation speed of the target point are affected by multiple factors and are fuzzy and uncertain. Therefore, the present invention uses interval numbers to represent uncertain parameters, and proposes an interval number dominance relationship based on interval possibility in the improved multi-objective genetic optimization method, which fully reflects the influence of the interval number, so that the model construction and solution process are more in line with the actual situation of equipment task scheduling. BRIEF DESCRIPTION OF THE DRAWINGS

[0071] Figure 1 It is the Pareto optimal solution obtained by using the improved genetic algorithm in the embodiment of the present invention.

[0072] Figure 2 It is a Gantt chart corresponding to a typical optimal task allocation solution 1 in an embodiment of the present invention.

[0073] Figure 3 It is a Gantt chart corresponding to the typical optimal task allocation scheme 2 in the embodiment of the present invention.

[0074] Figure 4It is a Gantt chart corresponding to the typical optimal task allocation scheme three in the embodiment of the present invention.

[0075] Figure 5 It is a Gantt chart corresponding to the typical optimal task allocation scheme 4 in the embodiment of the present invention.

[0076] Figure 6 It is a comparison between the algorithm of the present invention and the traditional NSGA-II algorithm in terms of coverage index.

[0077] Figure 7 It is a comparison between the algorithm of the present invention and the traditional NSGA-II algorithm in terms of uniformity index.

[0078] Figure 8 It is a comparison between the algorithm of the present invention and the traditional NSGA-II algorithm in terms of convergence index. DETAILED DESCRIPTION

[0079] The present invention provides a method for intelligent decomposition and scheduling of multiple equipment tasks based on target point clustering and improved genetic algorithm. The method can automatically decompose equipment tasks according to the characteristics of target points, and realize equipment-level task scheduling and timing optimization, thereby improving the automation and timeliness of equipment task decomposition and scheduling.

[0080] Specifically, the multi-equipment task intelligent decomposition and scheduling method based on target point clustering and improved genetic algorithm provided by the present invention comprises the following steps:

[0081] Step 1: Modeling equipment mission elements

[0082] The elements of equipment scheduling tasks include: equipment, tasks, target points and time. It is necessary to model the equipment, target points and time elements according to the actual equipment scheduling tasks. The target point is the location where the equipment performs a specific task.

[0083] Step 2: Equipment task decomposition based on target point clustering

[0084] In the process of equipment scheduling tasks, target points with close locations are often assigned to the same equipment group (a group of equipment that performs the same task at the same target point) to perform the corresponding tasks. Therefore, target points with close locations should be divided into the same group. In order to scientifically and reasonably group target points according to their locations, an intelligent clustering method for target point locations based on mean clustering is proposed. This method can automatically divide the target point locations into multiple clusters, so that the target point locations within each cluster are as close as possible, and the target point locations between clusters are as far as possible. The basic process of the intelligent clustering method for target point locations based on mean clustering is as follows.

[0085] Step 2.1: Read the location information of all target points and determine the number of clusters N to be divided K, randomly select N from all target point positions K The positions are taken as the initial clustering centers.

[0086] Step 2.2: For all target points, calculate their differences from these N K The distances between the cluster centers are used to divide them into the clusters with the closest cluster centers.

[0087] Step 2.3: In each newly divided cluster, calculate the average value of all positions in the cluster and use it as the new cluster center point.

[0088] Step 2.4: Repeat steps 2.2 and 2.3 until the clustering of the target point position no longer changes.

[0089] Step 3: Equipment task scheduling model construction

[0090] Based on the above step 2, after the target points are clustered, the tasks need to be decomposed according to the target point clusters and action types, namely:

[0091] For tasks of the same action type in the same target point cluster, they are considered as one task;

[0092] For tasks with different action types in the same target point cluster, they are considered as different tasks according to their action types;

[0093] For different target point clusters, whether they are the same action type or different action type tasks, they are considered as different tasks. Therefore, the number of tasks n in the equipment task set in this step is greater than the number of target point clusters N K .

[0094] Step 3.1: Equipment Capability Requirements Modeling

[0095] The task set of the equipment is denoted as T = {T1, T2, ···, T n}, n is the number of tasks in the set. Its properties include: task coordinate location (XT i ,YT i ); if it is the same type of task in the same cluster, the coordinates are the average; if it is different types of tasks in the same cluster, the coordinates are their respective coordinates; task duration TT i ;Task capability requirement vector (DT i1 ,DT i2 ,···,DT il ), where l is the number of different capability types, DT ik (1≤k≤l) is the size of the kth capability required to process task Ti. If DT ik =0, then process task T iThe kth capability is not required.

[0096] Step 3.2: Equipment Capability Modeling

[0097] The set of all equipment is denoted as A = {A1, A2, ···, A v}, v is the number of equipment. Its properties include: the initial coordinate position of the equipment (XA j ,YA j ); Equipment capability vector (OA j1 ,OA j2 ,...,OA jl ), which includes the equipment's movement speed VA j ; l is the number of different capability types, OA jk (1≤k≤l) is equipment A j The size of the kth capability, if OA jk =0, then equipment A j Does not have the kth ability.

[0098] The task allocation scheme of the equipment group can be represented by the matrix TAM=(tam ij ) n×v Indicates that if task T i Assigned to Equipment A j Processing, then tam ij =1, otherwise tam ij = 0. When a task cannot be processed because the conditions are not met, the equipment that has arrived at the task location will wait here until the task is completed.

[0099] Step 3.3: Modeling uncertainty parameters

[0100] In the process of equipment task allocation and scheduling, there are parameter uncertainties, including task duration TT i (1≤i≤n) uncertain, each component DT in the task capability requirement vector ik (1≤i≤n,1≤k≤l) Uncertain, equipment moving speed VA j (1≤j≤v) is uncertain and each component in the equipment capability vector (1≤j≤v,1≤k≤l) is uncertain. In a complex and ever-changing task execution environment, the distribution of these parameters is difficult to accurately count, but estimating their range is relatively simple, so interval numbers are introduced to characterize the uncertainty of these parameters.

[0101] Step 3.4: Modeling the timing of equipment task execution

[0102] There is usually a timing constraint relationship between the tasks in the task set T. The timing constraint relationship between the tasks in the task set T can be expressed by the matrix TCM = (tcmij ) n×n Indicates that if task T i It is task T j The predecessor task, then tcm ij =1, otherwise tcm ij = 0. A task can only be processed if it meets the following three conditions: first, all the predecessor tasks of the task have been completed; second, all the equipment assigned to handle the task has arrived at the location of the task; third, the various capabilities required to handle the task have been met to a certain extent.

[0103] The task priority is calculated using the weighted length algorithm (WL). In this algorithm, the priority coefficient of a task depends on the processing time of the task and the priority coefficient of the task's direct successor. The formula for calculating the task priority coefficient using the WL algorithm is as follows:

[0104]

[0105] Where WL(i) is the task T obtained by the WL algorithm. i The priority coefficient of task T. j The priority coefficient, T j It is T i The direct successor task, OUT(i) is the task sequence diagram T i The set of tasks that are the direct successors of . Indicates T i The maximum value of the task priority coefficient in the direct successor task set. j∈OUT(j) WL(j) represents T i The sum of the priority coefficients of all tasks in the direct successor task set.

[0106] Step 3.5: Modeling the equipment task scheduling problem

[0107] Time factors must be considered in equipment task scheduling, that is, The start time is recorded as BT i , the end time is recorded as ET i =BT i +TT i , the set of its predecessor tasks is denoted as ST Ti = {T k |tcm ki=1,1≤k≤n},in processing task T i When , the set of required capability types is denoted as R Ti ={k|DT ik >0, 1≤k≤u}, process T i The set of equipment is denoted as SA Ti = {A k|tam ik=1,1≤k≤u}.

[0108] When the equipment group performs various tasks according to the task allocation plan, the time to complete all tasks is

[0109] TM=max Ti∈T(ET i )

[0110] Among them: the smaller the TM, the shorter the time it takes for the equipment group to complete all tasks.

[0111] for Equipment Group Complete T i The quality is defined as

[0112]

[0113] in: T i The degree to which the kth capability requirement is satisfied, ||R Ti || is R Ti The number of elements in . The average quality of the equipment group to complete all tasks is defined as:

[0114]

[0115] Obviously, 0≤QM≤1, the larger the QM, the better the quality of the equipment group in completing all tasks.

[0116] The goal of equipment group task scheduling is to enable the entire group to complete all tasks in its task set T well and quickly, so the objective function of the task scheduling model is:

[0117] min TM; max QM

[0118] Considering 0≤QM≤1, the formula can be rewritten as

[0119] min f=(TM,1-QM)

[0120] Next, consider the constraints of the task scheduling model. If it is processing T i If no other tasks have been processed before, then it reaches task T i The time at your location is TIME ki =((XA k -XT i ) 2 +(YA k -YT i ) 2 ) 1 / 2 / VA k ; If it is processing T iThe last task processed before was T j , then it reaches T i The time at your location is TIME ki =((XT j -XT i ) 2 +(YT j -YT i ) 2 ) 1 / 2 / VA k +ET j When task T i To start being processed, the three conditions described above must be met, so there is

[0121]

[0122] QT i >0,

[0123] In summary, the equipment group task scheduling model is min f = (TM, 1-QM),

[0124]

[0125] Among them, tam ij is the decision variable of the model. From the above analysis, we can see that many parameters of the model are multi-objective optimization models that characterize uncertainty.

[0126] Step 4: Solving the equipment task scheduling model based on the improved multi-objective genetic algorithm

[0127] The goal of the model in step 3 is to minimize the time to complete all tasks and to achieve the best average quality of all tasks, that is, min TM = max Ti∈T(ET i ), This problem is a multi-objective optimization problem.

[0128] In order to improve the efficiency of the solution while ensuring the scientific nature of the solution, an interval-type non-dominated sorting algorithm is proposed based on the NSGA-II algorithm. Compared with NSGA-II, the present invention makes certain improvements to the NSGA-II algorithm process, and the improvements mainly include the following four points:

[0129] ① Propose a chromosome decoding method that satisfies the model constraints to ensure that the accessible space of the algorithm corresponds to the feasible domain of the model;

[0130] ② Define the dominance relationship of interval numbers based on interval possibility, which is used for comparison and ranking of model objectives;

[0131] ③ Use the interval distance formula to calculate the crowding distance of the solution, and use cyclic crowding sorting to eliminate redundant solutions;

[0132] ④ Design a heuristic crossover operator to guide population evolution and improve the convergence speed of the algorithm.

[0133] The algorithm solution process is divided into the following four steps:

[0134] Step 4.1: Chromosome encoding and decoding

[0135] Chromosome X is encoded in binary code, consisting of an n-row and u-column 0-1 matrix CM=(cm ij ) n×u Indicates that each row of the matrix CM is a gene of the chromosome. Due to the constraints of the established model, directly setting the task allocation scheme TAM to CM is not necessarily a feasible solution. Therefore, the chromosome needs to be decoded first. The specific steps of decoding are as follows:

[0136] Step 4.1.1: Initialize i=0;

[0137] Step 4.1.2: Set i=i+1. If i<n+1, set k=0 and execute step 4.1.3. Otherwise, the decoding process ends.

[0138] Step 4.1.3: Set k = k + 1. If k < l + 1, execute step 4.1.4. Otherwise, set j = 0 and execute step 4.1.5.

[0139] Step 4.1.4: If DT ik >0 and This means that when the task allocation scheme TAM = CM, the constraint QT i > 0 is not satisfied, and the solution is not a feasible solution. Therefore, it is necessary to find the equipment A with the kth ability and the smallest serial number value in the equipment set A. q , and let cm iq =1, then return to step 4.1.3, otherwise directly return to step 4.1.3;

[0140] Step 4.1.5: Let j = j + 1. If j < u + 1, execute step 4.1.6, otherwise return to step 4.1.2;

[0141] Step 4.1.6: If cm ij =1, execute step 4.1.7, otherwise return to step 4.1.5;

[0142] Step 4.1.7: Let TAM = CM, and calculate c1 = QT based on TAM i , then let tam ij = 0, here we calculate c2 = QT according to TAM i If c1=c2, it means that if task T iAssigned to Equipment A j If it is processed, it will not improve T i The quality of completion may also delay T i Therefore, in this case, let cm ij =0 and return to step 4.1.5, otherwise return directly to step 4.1.5.

[0143] The chromosome X=CM obtained after the decoding operation is a feasible solution of the model. At this time, CM can be set as the task allocation scheme TAM.

[0144] Step 4.2: Dominance relationship based on interval possibility

[0145] If the interval number A=[a L ,a R ],B=[b L , b R ], then the calculation formula of the interval-based probability P(A≤B) is:

[0146]

[0147] Considering that the probability P(A≤B) calculated by this formula has the following properties: ①0≤P(A≤B)≤1; ②P(A≤B)+P(B≤A)=1; ③If P(A≤B)=P(B≤A)=0.5, then a L =b L , a R =b R .

[0148] On this basis, the present invention defines the interval number dominance relationship based on interval possibility.

[0149] Definition 1: If the interval number A = [a L , a R ],B=[b L , b R ], then when P(A≤B)≥0.5, the interval number A is said to be no greater than the interval number B in the interval sense; when P(A≤B)<0.5, the interval number A is said to be greater than the interval number B in the interval sense; when P(A≤B)>0.5, the interval number A is said to be less than the interval number B in the interval sense; when P(A≤B)≤0.5, the interval number A is said to be no less than the interval number B in the interval sense.

[0150] Definition 2: If the objective function calculated after decoding chromosome X1 is TM1, 1-QM1, and the objective function calculated after decoding chromosome X2 is TM2, 1-QM2, then the following definitions apply:

[0151] ① When TM1, 1-QM1 is not greater than TM2, 1-QM2 in the interval sense, and TM1 is less than TM2 in the interval sense or 1-QM1 is less than 1-QM2 in the interval sense, then X1 is said to be superior to X2, and is recorded as X1>X2.

[0152] ② When TM1, 1-QM1 is not less than TM2, 1-QM2 in the interval sense, and TM1 is greater than TM2 in the interval sense or 1-QM1 is greater than 1-QM2 in the interval sense, then X1 is said to be inferior to X2, denoted as X1 <X2;

[0153] ③When X1 is neither better than X2 nor worse than X2, X1 is said to be indifferent from X2, and is recorded as X1~X2.

[0154] Definition 3: If the population P is a set of a certain number of chromosomes, X i ∈P, and when j≠i, there is no X j ∈P makes X1>X2 hold, then Xi is called a Pareto solution in population P.

[0155] Step 4.3: Sort by cyclic crowding distance

[0156] Due to the limited population size, when all chromosomes at a certain dominance level cannot enter the next generation population, it is necessary to calculate the crowding distance of these chromosomes and eliminate some dense solutions. Considering that the objective function TM and 1-QM are both interval numbers, the interval distance formula is required when calculating the crowding distance of chromosomes. The interval distance formula used in the present invention is:

[0157] d(A,B)=((a L +a R -b L -b R ) 2 / 4+((a R -a L ) 2 +(b R -b L ) 2 ) / 12-F 2 / 6) 1 / 2

[0158] Among them: A=[a L ,a R ],B=[b L , b R ]; F is the length of the interval A∩B, when When , F = 0. In addition, it should be pointed out that since the dimensions of the two objective functions TM and 1-QM are different, the objective functions need to be normalized before calculating the crowding distance.

[0159] In the original NSGA-II algorithm, the crowding distance of chromosomes is calculated once. The present invention improves on the basis of NSGA-II. For the calculation part of the crowding distance, the present invention selects chromosomes entering the next-generation population through cyclic crowding distance sorting, that is: first calculate the crowding distance of each alternative chromosome, then select the sparest (the largest crowding distance) chromosome into the next-generation population and delete it from the alternative chromosomes, and continuously repeat the above steps until the scale of the next-generation population reaches the upper limit. Although using cyclic crowding distance sorting increases the time complexity of the algorithm to a certain extent, it can better maintain the diversity of chromosomes in the population and avoid the algorithm from prematurely converging into a local optimum.

[0160] Step 4.4: Genetic operation

[0161] The genetic operators used in population evolution mainly include a selection operator, a mutation operator, and a crossover operator.

[0162] The selection operator is used to select chromosomes from the previous-generation population to perform crossover and mutation operations. The present invention adopts a binary tournament selection strategy.

[0163] The mutation operator generates new chromosomes through gene mutation. The present invention selects the mutation operator to adopt a uniform mutation strategy.

[0164] The crossover operator generates new chromosomes by recombining the genes of a pair of chromosomes. The crossover operator of the present invention is divided into two types: a random crossover operator and a heuristic crossover operator. Among them, the random crossover operator adopts a uniform crossover strategy, and the way the heuristic crossover operator constructs new chromosomes is as follows.

[0165] Step 4.4.1: Select the chromosomes participating in crossover X1 = CM1, X2 = CM2, initialize the new chromosome X3 = CM3, where all elements of the matrix CM3 are 1, and set i = 1.

[0166] Step 4.4.2: If i < n + 1, then execute Step 4.4.3, otherwise end and output X3 as the constructed new chromosome.

[0167] Step 4.4.3: Replace the i-th gene of X3 with the i-th genes of X1 and X2 respectively, and denote the formed new chromosomes as X 31 , X 32 . If X 31 > X 32 , then let X3 = X 31 ; if X 31 < X 32 , then let X3 = X 32 ; if X 31 ~X 32, then from X 31 , X 32 Randomly select an update X3 from . Set i=i+1 and return to step 4.4.2.

[0168] The random crossover operator essentially performs a random search in the solution space, which can easily lead to a slow convergence of the algorithm. The introduction of the heuristic crossover operator can continuously obtain some better solutions, thereby guiding the search in some more favorable directions, thereby improving the convergence speed of the algorithm. The combination of these two crossover operators is as follows: set the heuristic operator trigger rate ε (0≤ε≤1), and when the crossover operator is needed, generate a random number δ in the interval [0,1]. If δ≤ε, use the heuristic crossover operator, otherwise use the random crossover operator.

[0169] The method of the present invention is described below with reference to specific cases.

[0170] In the emergency repair task of multiple roads under the background of a natural disaster, the multi-equipment task intelligent scheduling method proposed above was comprehensively applied to the decomposition and scheduling problems of the emergency repair tasks of multiple roads. The implementation process and results are as follows.

[0171] First, the repair points are clustered and grouped according to their locations. Then, the road emergency repair tasks are decomposed according to the repair point groups and repair task types, so that repair points close to each other are assigned to the same equipment group to perform emergency repair tasks as much as possible. The road emergency repair task is finally decomposed into T1~T 18 There are 18 subtasks in total.

[0172] There are 8 possible capability requirements for road emergency repair tasks (DT i1 ,DT i2 ,DT i3 ,DT i4 ,DT i5 ,DT i6 ,DT i7 ,DT i8 ). According to the capability requirements of each subtask, the task attribute table is shown in Table 1:

[0173] Table 1 Task attribute table

[0174]

[0175]

[0176] The available repair equipment is A1~A20, and the equipment parameters are shown in Table 2:

[0177] Table 2 Equipment capability parameters

[0178]

[0179]

[0180] Considering the uncertainty of mission capability requirements and equipment capabilities, the attributes of mission and intelligence after the reference interval are shown in Table 3 and Table 4 respectively.

[0181] Table 3 Task attribute table represented by interval

[0182]

[0183]

[0184] Table 4 Equipment attribute table represented by interval

[0185]

[0186]

[0187] In order to verify the rationality and scientificity of the scheduling model and algorithm proposed in this invention, the model and algorithm are implemented through matlab code. In the implementation of task resource allocation and scheduling using the improved genetic algorithm, the numerical settings of the algorithm are as follows: the upper limit of the evolutionary generation g is set to 200, the population size w is set to 100, and the crossover probability p is set to 100. c Set to 0.5, the mutation probability p m is set to 0.05, and the heuristic operator trigger rate ε is set to 0.05.

[0188] The experimental results are as follows Figure 1 As shown in the figure, each rectangle represents a Pareto optimal solution. From the result graph, we can see that the algorithm obtains a set of Pareto solutions with relatively uniform distribution, which means that the algorithm is feasible and can be used to solve this type of uncertain multi-objective optimization model with interval parameters.

[0189] For multi-objective optimization problems, the Pareto optimal solution is a set of solutions that can be selected by the commander. The typical optimal task allocation solutions obtained by this algorithm are shown in Tables 5 to 8. The scheduling Gantt charts corresponding to these solution results are shown in Tables 5 to 8. Figure 2-5 shown.

[0190] Table 5 Typical optimal task allocation scheme 1

[0191]

[0192]

[0193] Table 6 Typical optimal task allocation scheme 2

[0194]

[0195] Table 7 Typical task allocation scheme 3

[0196]

[0197]

[0198] Table 8 Typical optimal task allocation scheme 4

[0199]

[0200]

[0201] In order to further demonstrate the feasibility and superiority of the algorithm, the improved algorithm is compared with the original NSGA-II algorithm. The coverage index α, convergence index β and uniformity index γ are used to measure the performance of the algorithm. Among them, the coverage index can be used to measure the dispersion of the Pareto frontier, and the larger the index, the better. The convergence index represents the distance between the approximate Pareto optimal solution set obtained by the algorithm and the ideal Pareto frontier. The smaller the index, the better the convergence of the solution obtained by the algorithm, and the closer it is to the ideal Pareto frontier. The smaller the uniformity index, the better, which means that the distribution of the obtained solution is more uniform. Based on the interval distance formula d(A,B), the formulas for obtaining the three indicators are as follows, where n is the number of Pareto solutions:

[0202]

[0203] in,

[0204]

[0205] In order to eliminate the influence of different optimization target dimensions, the objective function is normalized and then the three algorithm performance evaluation indicators are calculated respectively. Figures 6 to 8 shown.

[0206] Figures 6 to 8 In the figure, the dotted line is the result corresponding to the traditional NSGA-II, and the solid line is the result of the algorithm improved based on the NSGA-II algorithm of the present invention. From the analysis of the three experimental result graphs, it can be seen that the Pareto solution set obtained by the algorithm of the present invention has achieved good results in terms of convergence, uniformity and coverage.

Claims

1. A multi-equipment task intelligent decomposition and scheduling method based on target point clustering and improved genetic algorithm, characterized in that: The steps include: Step S1, modeling the elements of a multi-equipment task; the elements include equipment, tasks, target points and time; Step S2, decomposing the multi-equipment tasks based on target point clustering; Step S3, constructing a multi-equipment task scheduling model; Step S4, solving the multi-equipment task scheduling model based on the improved multi-objective genetic algorithm; Step S4 is specifically as follows: Step S4-1: adopting a chromosome decoding method that satisfies the model constraints to ensure that the reachable space of the algorithm corresponds to the feasible domain of the model; Step S4-2: The dominance relationship of the number of intervals based on the interval possibility is used for comparison and ranking of model targets; Step S4-3: Calculate the crowding distance of the solution using the interval distance formula, and use cyclic crowding sorting to eliminate redundant solutions; Step S4-4: Design a heuristic crossover operator to guide population evolution, improve the convergence speed of the algorithm, and ultimately obtain a solution to multi-equipment task scheduling.

2. The method for intelligent decomposition and scheduling of multi-equipment tasks based on target point clustering and improved genetic algorithm according to claim 1 is characterized in that: Step S2 is specifically as follows: Step S2-1: Determine the location information of all target points and the number of clusters N to be divided K , randomly select N from all target point positions K The positions are taken as the initial clustering centers; Step S2-2: For all target points, calculate their differences from these N K The distance between the cluster centers is divided into clusters with the closest cluster center. Step S2-3: In each newly divided cluster, calculate the average value of all positions in the cluster and use it as the new cluster center point; Step S2-4: Repeat steps S2-2 and S2-3 until the clustering of the target point position no longer changes.

3. The method for intelligent decomposition and scheduling of multi-equipment tasks based on target point clustering and improved genetic algorithm according to claim 1 is characterized in that: Step S3 is as follows: Step S3-1: Model the equipment capability requirements; record the equipment task set as T = {T1, T2, ···, T n }, n is the number of tasks in the set; for Its properties include: task coordinate location (XT i ,YT i ); Task duration TT i ;Task capability requirement vector (DT i1 ,DT i2 ,···,DT il );1≤k≤l; Step S3-2: Model the equipment capability; the set of all equipment is denoted as A = {A1, A2, ···, A v }, v is the number of equipment; for Its properties include: the initial coordinate position of the equipment (XA j ,YA j ), equipment capability vector (OA j1 ,OA j2 ,...,OA jl ), the equipment's movement speed VA j ; 1≤k≤l; The task allocation scheme of the equipment group is represented by the matrix TAM=(tam ij ) n×v Indicates that if task T i Assigned to Equipment A j Processing, then tam ij =1, otherwise tam ij =0; Step S3-3: Modeling interval uncertainty parameters for equipment capability requirements and equipment capabilities; Step S3-4: Model the execution sequence of multiple equipment tasks; the timing constraint relationship between tasks in the task set T is represented by the matrix TCM=(tcm ij ) n×n Indicates; if task T i It is task T j The predecessor task, then tcm ij =1, otherwise tcm ij =0; The task priority is calculated using the weighted length algorithm WL. The specific formula is as follows: Where WL(i) is the task T obtained by the WL algorithm. i The priority coefficient of task T j The priority coefficient, T j It is T i The direct successor task, OUT(i) is the task sequence diagram T i The set of tasks that are the direct successors of Indicates T i The maximum value of the task priority coefficient in the direct successor task set, ∑ j∈OUT(j) WL(j) represents T i The sum of the priority coefficients of all tasks in the direct successor task set; Step S3-5: Model the multi-equipment task scheduling based on the task completion time and completion quality.

4. The method for intelligent decomposition and scheduling of multiple equipment tasks based on target point clustering and improved genetic algorithm according to claim 3 is characterized in that: Step S3-5 is as follows: for The start time is recorded as BT i , the end time is recorded as ET i =BT i +TT i , the set of its predecessor tasks is denoted as ST Ti = {T k |tcm ki=1,1≤k≤n},in processing task T i When , the set of required capability types is denoted as R Ti ={k|DT ik >0, 1≤k≤u}, process T i The set of equipment is denoted as SA Ti = {A k |tam ik=1,1≤k≤u}; The objective function of the multi-equipment task scheduling model is min TM; max QM TM is the time to complete all tasks, QM is the quality of completing all tasks; for Ti, if it is processing T i If no other tasks have been processed before, then it reaches task T i The time at your location is TIME ki =((XA k -XT i ) 2 +(YA k -YT i ) 2 ) 1 / 2 / VA k ; If it is processing T i The last task processed before was T j , then it reaches T i The time at your location is TIME ki =((XT j -XT i ) 2 +(YT j -YT i ) 2 ) 1 / 2 / VA k +ET j ; The multi-equipment task scheduling model is min f = (TM, 1-QM), Among them, tam ij is the decision variable of the model.

5. The method for intelligent decomposition and scheduling of multi-equipment tasks based on target point clustering and improved genetic algorithm according to claim 4 is characterized in that: Step S4-1 is specifically as follows: Chromosome X is encoded in binary code, consisting of an n-row and u-column 0-1 matrix CM=(cm ij ) n×u It means that each row of the matrix CM is a gene of the chromosome; Step S4-1-1: Initialize i=0; Step S4-1-2: Set i=i+1. If i<n+1, set k=0 and execute step S4-1-3. Otherwise, the decoding process ends. Step S4-1-3: Set k=k+1. If k<l+1, execute step S4-1-4. Otherwise, set j=0 and execute step S4-1-5. Step S4-1-4: If DT ik >0 and This means that when the task allocation scheme TAM = CM, the constraint QT i > 0 is not satisfied, and it is necessary to find the equipment A with the kth ability and the smallest serial number value in the equipment set A. q , and let cm iq =1, then return to step S4-1-3, otherwise directly return to step S4-1-3; Step S4-1-5: Let j=j+1. If j<u+1, execute step S4-1-6, otherwise return to step S4-1-2; Step S4-1-6: If cm ij =1, execute step S4-1-7, otherwise return to step S4-1-5; Step S4-1-7: Let TAM = CM, and calculate c1 = QT based on TAM i Then let tam ij = 0, and calculate c2 = QT according to TAM i ; If c1 = c2, let cm ij =0 and return to step S4-1-5, otherwise directly return to step S4-1-5; The chromosome X=CM obtained after the decoding operation is a feasible solution of the model. At this time, CM is set as the task allocation scheme TAM.

6. The method for intelligent decomposition and scheduling of multiple equipment tasks based on target point clustering and improved genetic algorithm according to claim 5 is characterized in that: In step S4-3, the chromosomes to be entered into the next generation population are selected by cyclic crowding distance sorting, that is, firstly, the crowding distance of each candidate chromosome is calculated, then the chromosome with the largest crowding distance is selected into the next generation population and deleted from the candidate chromosomes, and the above steps are repeated until the size of the next generation population reaches the upper limit; The interval distance formula used to calculate the crowded distance of chromosomes is: d(A,B)=((a L +a R -b L -b R ) 2 / 4+((a R -a L ) 2 +(b R -b L ) 2 ) / 12-F 2 / 6) 1 / 2 Among them: A=[a L ,a R ],B=[b L , b R ]; F is the length of the interval A∩B, when When , F = 0; the two objective functions TM and 1-QM have different dimensions, so the objective function needs to be normalized before calculating the crowding distance.

7. The method for intelligent decomposition and scheduling of multiple equipment tasks based on target point clustering and improved genetic algorithm according to claim 6 is characterized in that: Step S4-4 is specifically as follows: Step S4-4-1: Select the chromosomes X1 = CM1 and X2 = CM2 participating in the crossover, initialize the new chromosome X3 = CM3, where all elements of the matrix CM3 are 1, and set i = 1; Step S4-4-2: If i < n + 1, then execute Step S4-4-3, otherwise end and output X3 as the constructed new chromosome; Step S4-4-3: Replace the i-th gene of X3 with the i-th gene of X1 and X2 respectively, and the resulting new chromosomes are denoted as X 31 , X 32 ; if X 31 >X 32 , then let X3=X 31 ; if X 31 <X 32 , then let X3=X 32 ; if X 31 ~X 32 , then from X 31 , X 32 Randomly select an update X3; let i=i+1 and return to step S4-4-2.

Images