Multi-target optimization method oriented to multi-agricultural machinery collaborative scheduling

Abstract

The invention discloses a multi-target optimization method oriented to multi-agricultural-machinery collaborative scheduling, and relates to the technical field of scheduling optimization and agricultural intelligent scheduling. According to the method, a multi-objective optimization model with the goal of minimizing the maximum completion time of all farmland tasks and the total transfer time of agricultural machinery is constructed, constraints such as path integrity and task coverage are met, the agricultural machinery is clustered, divided and distributed through a clustering algorithm according to farmland space distribution and task load, and an initial population is constructed; a scheduling scheme is optimized by adopting a two-stage evolution mechanism, a path is optimized in a cluster at the initial stage, a path is reconstructed in a cross-cluster manner at the later stage, and load balancing task migration is implemented; and simulating a collaborative operation process through an event-driven decoding mechanism, and outputting a scheduling scheme. The method can dynamically describe the agricultural machinery collaborative operation process, improves the scheduling efficiency and balance, and is suitable for a multi-agricultural machinery collaborative scheduling scene.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of scheduling optimization and agricultural intelligent scheduling, in particular to a multi-objective optimization method for multi-tractor collaborative scheduling. BACKGROUND

[0002] With the continuous advancement of agricultural scale and intelligentization, tractors play an irreplaceable role in plowing, sowing, harvesting and other links. In actual operation process, how to scientifically and reasonably schedule multiple tractors has become a key problem to ensure the efficiency of agricultural production and improve the utilization rate of resources.

[0003] In actual scheduling process, multiple mutually restrictive optimization objectives such as task completion time, transfer cost, operation rate and load balancing are usually involved, and there is a strong trade-off relationship between each objective. The multi-objective scheduling model can reasonably allocate tasks and plan paths by coordinating the collaborative scheduling relationship of multiple tractors, providing efficient and implementable operation scheme for tractor organization. Research on multi-objective tractor collaborative scheduling problem for actual demand has important theoretical and practical significance for efficient allocation of tractor resources and improvement of agricultural intelligent level. This kind of multi-tractor collaborative scheduling multi-objective optimization problem exists widely in practical application, with many decision variables, complex coupling relationship and large calculation space, which belongs to typical NP-hard problem. In addition, the actual tractor scheduling process has natural dynamic characteristics and collaborative challenges:

[0004] On the one hand, the collaborative operation efficiency of tractors arriving at the same task point at different times presents dynamic changes, and the operation progress or operation completion time is difficult to estimate statically; on the other hand, the optimization of operation path not only depends on the geographical distribution of farmland, but also is affected by many factors such as farmland task quantity and tractor performance; at the same time, in the scheduling process, if the task load is not evenly distributed, it will lead to overloading of some tractors and idling of some tractors, which is easy to cause uneven resource utilization and increase equipment failure rate.

[0005] At present, the tractor scheduling model is usually modeled as vehicle routing problem (VRP) or multi-robot task allocation (MRTA) model, which has certain application value in general logistics and manufacturing scenarios, but still has significant deficiencies in dealing with this kind of complex problem with collaboration and dynamics, which is manifested in:

[0006] 1. The traditional model generally assumes a "single tractor single field" operation scenario in which a piece of farmland is completed by a single tractor independently, which cannot reflect the collaborative scenario of "subsequent tractor joining the unfinished task" in actual production;

[0007] 2. In traditional MRTA or VRP frameworks, task completion time is usually set as a static value, the duration of the job does not change with the actual job process, and the task completion is usually based on the completion of the last device, lacking dynamic expression of the collaborative process.

[0008] Furthermore, from an algorithmic perspective, current scheduling algorithms do not adequately consider the problem of coordinated scheduling of multiple agricultural machines, which is manifested in the following ways:

[0009] 1. Insufficient consideration of the load balancing problem of agricultural machinery can easily lead to overload operation of some agricultural machinery, affecting the continuity of operation and overall efficiency;

[0010] 2. When faced with large-scale instances, the algorithm's convergence speed is insufficient, affecting actual scheduling efficiency.

[0011] Therefore, a more practically adaptable scheduling optimization model and algorithm are needed. The model should possess the ability to realistically depict the dynamic process of collaborative operations and adapt to multi-objective optimization requirements. The algorithm should accelerate convergence in large-scale instances, alleviate resource underutilization caused by uneven job allocation, and meet the scheduling requirements for high-efficiency development. Summary of the Invention

[0012] To address the shortcomings of existing models in effectively depicting the dynamic collaborative relationships among agricultural machinery during task execution, and the insufficient consideration of load balancing and convergence speed in scheduling algorithms, this invention proposes a multi-objective optimization method for multi-machine collaborative scheduling. At the modeling level, it supports simultaneous operation of multiple machines on the same farmland and introduces a "task volume - collaborative operation rate" integral mechanism to dynamically characterize the dynamic coupling relationship between collaborative operation rate and task volume. In terms of algorithm design, a two-stage multi-objective optimization algorithm (C2LEA algorithm) integrating clustering guidance and a load-balancing-based task migration strategy is proposed. Furthermore, an event-driven decoding mechanism is constructed to record the operation status in real time based on event tuples, enabling dynamic simulation of the scheduling scheme. The research focuses on optimizing the maximum task completion time and the total machine transfer time, providing theoretical support and feasible solutions for agricultural machinery collaborative scheduling.

[0013] The technical solution adopted in this invention is as follows:

[0014] A multi-objective optimization method for coordinated scheduling of multiple agricultural machines includes the following steps:

[0015] Step 1, construct a system to minimize the maximum completion time for all farmland tasks: And minimize the total transfer time of all agricultural machinery: A multi-objective optimization model with the objective of [objective name missing].

[0016] Where N represents the number of farmlands; M represents the number of agricultural machines; t ij represents the transition time from farmland i to task point j; represents whether the kth agricultural machine performs the transition from farmland i to farmland j, 1 represents execution, and 0 represents non-execution; ct i represents the completion time of farmland i;

[0017] Step 2, a clustering algorithm is used to cluster according to the spatial distribution and task quantity of farmland tasks, and then the number of agricultural machines is allocated according to the cluster size; subsequently, an initial task configuration is generated by randomly allocating farmland tasks of the cluster to each agricultural machine, so as to construct an initial population of the algorithm; each individual in the population is encoded by using a real number vector to represent the operation scheduling scheme of the agricultural machine, and each individual code is a two-dimensional real number matrix, the row represents the agricultural machine number, and the column represents the farmland index accessed by the corresponding agricultural machine, constituting the operation sequence of the agricultural machine;

[0018] Step 3, the scheduling scheme is optimized based on a two-stage evolution mechanism:

[0019] Early stage: cross and local search operations of the agricultural machine path are performed within the cluster;

[0020] Late stage: the agricultural machine path is reconstructed across the cluster, and a task migration strategy based on load balancing is triggered;

[0021] Step 4, the agricultural machine cooperative operation process is accurately simulated through an event-driven decoding mechanism, and the scheduling scheme is output.

[0022] Further, the multi-objective agricultural machine scheduling optimization model satisfies the following constraints:

[0023] In-out balance constraint, each agricultural machine must have a clear "arrival" and "departure" operation for each farmland it visits in its scheduling path, which is expressed by the formula:

[0024]

[0025] Task coverage constraint, the problem solution needs to ensure that all farmlands are completely worked by at least one agricultural machine, that is, there is no uncovered task point in the system, and the formula is expressed as:

[0026]

[0027] Operation non-repetition constraint, the same agricultural machine cannot perform multiple operation operations on the same farmland; the agricultural machine can only transfer to the next task point after the target farmland task quantity is zero, which is expressed by the formula:

[0028]

[0029] The dispatch starting point normalization constraint is that the starting state of all agricultural machines is set to start from the dispatch center, and the initial arrival time and the starting time are both set to 0, as the unified time origin of the dispatch process, which is expressed as:

[0030]

[0031] Wherein, at 0k represents the time when the kth agricultural machine starts from the garage; ct0 represents that the garage completion time is set to 0, i.e. does not need to be completed, and index 0 represents the garage or the dispatch center;

[0032] The time reachability constraint is that the time when each agricultural machine arrives at the next task point should be equal to the sum of the work completion time of the previous task point and the moving time between the two points, which is expressed as:

[0033]

[0034] Wherein, at jk represents the time when the kth agricultural machine arrives at the farmland j;

[0035] The cooperative work rate constraint is that the cooperative work rate of the farmland at any time is the superposition of the work rates of the agricultural machines present at that time, which is expressed as:

[0036]

[0037] Wherein, E j (t) is the cooperative work rate of the farmland j at any time t, M j (t) represents the set of agricultural machines that cooperatively work on the farmland j at time t; e k represents the work rate of the kth agricultural machine;

[0038] The cooperative work completion time constraint is that the completion time of the farmland task is determined by the integral relationship between the total demand and the cooperative work rate, and it is ensured that the task amount of each farmland needs to be completed, which is expressed as:

[0039]

[0040] Wherein, Q j represents the task amount of the farmland j;

[0041] The agricultural machine work duration constraint is that the work duration of each agricultural machine on the farmland task point is obtained by the difference between the completion time of the farmland task and the actual arrival time of the agricultural machine, which is expressed as:

[0042]

[0043] Wherein, τ jkThe operation time of the kth agricultural machine on the farmland j;

[0044] The variable domain constraint is used to describe that the decision variables of the scheduling path and the task allocation should satisfy the 0-1 binary requirement, and the formula is expressed as:

[0045] Further, the step 2 specifically comprises the following steps:

[0046] Step 2-1, based on the spatial coordinates (x i ,y i ) of the farmland and the farmland task quantity Q j , a clustering feature vector is constructed; a K-Means clustering algorithm is used to perform spatial clustering on all farmland points, and the clustering number k is iterated in the range of 2 to "farmland number-1", and the optimal clustering division is selected in combination with the silhouette coefficient; wherein the silhouette coefficient is used as a clustering effect evaluation index, which can measure the intra-cluster compactness and inter-cluster separation at the same time, and the larger the value is, the better the clustering effect is; the calculation formula is as follows:

[0047]

[0048] Wherein, a(i) represents the average distance from sample i to other samples in the same cluster, and b(i) represents the average distance from sample i to all samples in the nearest neighbor cluster.

[0049] Step 2-2, according to the relative proportion of the total amount of tasks in each cluster, a corresponding number of agricultural machine resources is allocated to each cluster, and each agricultural machine is randomly allocated the agricultural machine task of its own cluster to form an initial task configuration scheme, and an initial population of the algorithm is constructed; wherein each individual in the population is encoded by a real number vector to represent the operation scheduling scheme of the agricultural machine, and each individual code is a two-dimensional real number matrix, the behavior agricultural machine number, and the list represents the farmland index accessed by the corresponding agricultural machine, forming the operation order of the agricultural machine.

[0050] Further, the step 3 specifically comprises the following steps:

[0051] Step 3-1, in the initial stage, local search and intra-cluster individual crossover are performed on each cluster; local search and local reordering are used to improve the convergence speed, and a high-quality initial solution basis is provided for subsequent global scheduling optimization;

[0052] Step 3-2, in the later stage, based on the initial solution set, cross-cluster crossover and mutation are performed to allow cross-cluster operation path reconstruction; at the same time, a load balancing migration strategy is introduced to consider the possibility of cross-cluster collaborative operation.

[0053] Further, the step 3-1 specifically comprises the following steps:

[0054] Step 3-1-1, according to the cluster label in step 2, select the same cluster of agricultural machinery;

[0055] Step 3-1-2, for the operation path of the agricultural machinery in the same cluster, perform crossover operation using OX and PMX crossover operators, and perform local search using path reversal, path insertion and path insertion three types of local search operators, and on this basis, optimize the intra-cluster operation sequence.

[0056] Further, the step 3-2 specifically includes the following steps:

[0057] Step 3-2-1, cancel the cluster label restriction, and allow different cluster agricultural machinery to cross-cluster crossover and variation;

[0058] Step 3-2-2, perform crossover operation on the operation paths of different cluster agricultural machinery, and perform cross-cluster cluster operation path reconstruction using three types of local search operators (see step 3-1-2);

[0059] Step 3-2-3, based on the current population individuals, perform load balancing judgment: calculate the load balancing degree coefficient cv of each agricultural machinery operation time: the ratio of the standard deviation and the average value of the working time of each agricultural machinery, and the calculation formula is as follows:

[0060]

[0061] Where std{workTime_List} and mean{workTime_List} represent the standard deviation and the average value of the working time of the agricultural machinery, respectively; if cv < δ, the load is relatively balanced, and the number of migrated agricultural machinery is selected as 1 / 8 of the total number of agricultural machinery; if cv > δ, it is considered that the load is relatively unbalanced, and the number of migrated agricultural machinery is selected as 1 / 4 of the total number of agricultural machinery; δ is the load balancing coefficient;

[0062] Step 3-2-4, arrange the agricultural machinery in ascending order of operation time, select the front number of migrated agricultural machinery as the low-load agricultural machinery, and the rear number of migrated agricultural machinery as the high-load agricultural machinery;

[0063] Step 3-2-5, for each high-load agricultural machinery, select the longest task in its task list as the migration candidate set;

[0064] Step 3-2-6, traverse the possible insertion positions of the low-load agricultural machinery to generate a migration candidate solution; select the optimal migration scheme through weighted comprehensive scoring; the migration score calculation formula is as follows:

[0065] Score = α·ΔR total + β·ΔT max + γ·ΔVar

[0066]

[0067] Wherein, σ orign , σ new respectively represent total transfer time before and after migration, maximum completion time before and after migration, and agricultural machinery load variance before and after migration; alpha, beta and gamma are weight coefficients, which can be set according to actual requirements or decision preferences.

[0068] Further, the step 4 specifically comprises the following steps:

[0069] Step 4-1, a real number vector coding is used to represent the operation scheduling scheme of agricultural machinery, each decoding individual is a two-dimensional real number matrix, the row represents the agricultural machinery number, and the column represents the agricultural field index accessed by the corresponding agricultural machinery, thereby constituting the operation sequence of the agricultural machinery;

[0070] Step 4-2, event initialization and queue construction: according to the individual coding result, the moving time to the first agricultural field is calculated, the initial "arrival event" is generated and added to the event queue;

[0071] Step 4-3, arrival event processing: when a certain agricultural machinery arrives at the target agricultural field, the system first updates the remaining operation amount of the task point, identifies the set of cooperative agricultural machinery currently in the agricultural field, calculates the cooperative operation rate and the task completion time (the departure time of the cooperative operation agricultural machinery) according to the cooperative operation rate, and updates the agricultural machinery state to "in operation"; departure event processing: when the remaining operation amount of the agricultural field task point is zero, the task is considered to be completed; the actual operation time, task state and completion time of the cooperative agricultural machinery are updated, and the corresponding departure event is generated and added to the event sequence; at the same time, the agricultural machinery is scheduled to the next agricultural field task point in its task sequence, the arrival time is calculated, and the state is updated to "in transition";

[0072] Step 4-4, loop decoding process: the decoding process continuously advances in an event-driven manner; the system selects the agricultural machinery with the earliest next event occurrence time by searching the scheduled departure time of the agricultural machinery currently in the "in operation" state and the scheduled arrival time of the agricultural machinery in the "in transition" state, and judges the event type according to the state: if the state is "in operation", the "departure event" is triggered, and the task departure processing logic is executed; if the state is "in transition", the "arrival event" is triggered, and the task arrival processing logic is executed; the decoding is continuously performed until all tasks are completed.

[0073] Step 4-5, output event tuple list, the events are arranged in chronological order, and the cumulative working time and two target values of each agricultural machinery are output; wherein each event tuple comprises the agricultural machinery number, the agricultural field number, the event time and the event type.

[0074] The present application has the beneficial effects of:

[0075] At the modeling level, the application breaks through the limitations of traditional models and constructs a more practical multi-objective optimization model. The traditional model generally assumes that a piece of farmland is only completed by a single agricultural machine independently, and the task completion time is mostly a static value, which cannot reflect the collaborative scenario of "subsequent agricultural machines joining the unfinished task" and the dynamic changes of the operation process. While the model of the application supports multiple agricultural machines working in the same farmland simultaneously, the collaborative operation rate is dynamically superimposed with the number of participating agricultural machines and time, and the "task volume-collaborative operation rate" integral mechanism is introduced, so that the operation time is dynamically adjusted with the arrival and departure of agricultural machines, and the task completion speed is determined by the rate of real-time on-site agricultural machines, which can accurately depict the dynamic process of collaborative operation.

[0076] In terms of algorithm design, the application adopts a two-stage multi-objective optimization algorithm that integrates clustering guidance and task migration strategy based on load balancing, effectively improving the scheduling efficiency and solution quality. By constructing a feature vector based on the spatial coordinates of the farmland and the task volume, the optimal cluster number is adaptively determined by K-Means clustering combined with the contour coefficient, and the agricultural machine resources are allocated according to the cluster, reducing the invalid movement of agricultural machines in the farmland, providing a high-quality initial solution for the algorithm, and improving the convergence speed. In the initial stage, cross and local search are performed within the cluster to quickly optimize the operation sequence within the cluster; in the later stage, the path is reconstructed across the clusters, and a load-balanced task migration strategy is introduced, which dynamically adjusts the number of migrated agricultural machines by calculating the load balancing coefficient, balances the operation time of each agricultural machine, avoids overloading of some agricultural machines, and further optimizes the scheduling scheme.

[0077] In practical application, the event-driven decoding mechanism records the operation state in real time through event tuples, realizes the dynamic simulation of the scheduling scheme, and ensures the feasibility of the scheme. Experimental comparison shows that, compared with ant colony algorithm, meme algorithm, NSGA-II, etc., the algorithm of the application performs better in IGD (convergence) and HV (solution set diversity) indicators, especially on medium and large-scale data sets, and can obtain a solution set with better convergence and more uniform distribution. At the same time, the algorithm effectively reduces the total transfer cost, which is about 25% lower than the traditional algorithm on average, and reduces the risk of agricultural machine failure caused by overload through the load balancing strategy, improves the operation continuity and overall efficiency, provides a practical and feasible scheme for multi-agricultural machine collaborative scheduling, and is suitable for intelligent operation management needs in fragmented farmland areas and large field environments. BRIEF DESCRIPTION OF DRAWINGS

[0078] Figure 1 Schematic diagram for multi-agricultural machine collaborative scheduling;

[0079] Figure 2 Farmland clustering result graph on the data set of the application;

[0080] Figure 3 Encoding display graph of the application;

[0081] Figure 4 A method flowchart of the present application;

[0082] Figure 5 A scheduling cost comparison chart of four methods on six data sets of the present application;

[0083] Figure 6 A scheduling Gantt chart of the present application

[0084] Figure 7 A dynamic collaborative work process schematic diagram of the present application;

[0085] Figure 8 An ablation experiment result comparison chart of the present application. DETAILED DESCRIPTION

[0086] The present application will be further described below in conjunction with the accompanying drawings.

[0087] The present application is a multi-objective optimization method for multi-tractor collaborative scheduling, which is used to solve the efficient collaborative scheduling problem of multiple tractors in multiple farmland tasks. In the scheduling process, the system needs to give an optimal scheduling scheme according to the input conditions such as farmland distribution position, task quantity, tractor performance parameters and initial position of the scheduling center, so as to realize the minimization of the collaborative work completion time of the tractors and the minimization of the total transfer time of the tractors, as shown in the following formula: Figure 1

[0088] The present application comprises the following steps:

[0089] Step 1: In order to solve the problem that the traditional model generally adopts the "one tractor one field" work assumption, which is difficult to effectively depict the dynamic collaborative relationship of the tractors in the task execution process, and the actual scheduling has multiple objectives to be considered, the present application takes the minimization of the maximum completion time of all farmland tasks: and the minimization of the total transfer time of all tractors: as the optimization objective, and constructs a multi-objective optimization model; allows multiple tractors to work in the same farmland, and efficiently restores the work process according to the mathematical model in the decoding, considers that the work efficiency and the task completion time change with the arrival time of the tractors and the number of working tractors, and realizes the dynamic collaboration of the scheduling model.

[0090] Wherein, N represents the number of farmlands; M represents the number of tractors; t ij represents the transfer time from farmland i to task point j; represents whether the kth tractor is from farmland i to farmland j, 1 represents execution, and 0 represents non-execution; ct i represents the completion time of farmland i.

[0091] The multi-objective tractor scheduling optimization model in the present application satisfies the following constraints: ​

[0092] In-out balance constraint, each agricultural machine must have a clear "arrival" and "departure" operation for each farmland it visits in its scheduling path, which is expressed as:

[0093]

[0094] Task coverage constraint, the problem solution needs to ensure that all farmlands are completely worked by at least one agricultural machine, that is, there is no uncovered task point in the system, and the formula is expressed as:

[0095]

[0096] Non-repetition of operation constraint, the same agricultural machine cannot perform multiple operation operations on the same farmland; the agricultural machine can only transfer to the next task point after the target farmland task quantity is zero, and the formula is expressed as:

[0097]

[0098] Scheduling starting point normalization constraint, the scheduling starting state of all agricultural machines is set to start from the scheduling center, and the initial arrival time and start time are set to 0 as the unified time origin of the scheduling process, and the formula is expressed as:

[0099]

[0100] Where, at 0k represents the time when the kth agricultural machine leaves the garage; ct0represents that the garage completion time is set to 0, that is, it does not need to be completed, and index 0 represents the garage or scheduling center;

[0101] Time accessibility constraint, the time when each agricultural machine arrives at the next task point should be equal to the sum of the operation completion time of the previous task point and the moving time between the two points, and the formula is expressed as:

[0102]

[0103] Where, at jk represents the time when the kth agricultural machine arrives at farmland j;

[0104] Cooperative operation rate constraint, the cooperative operation rate of a farmland at any time is the superposition of the operation rates of the agricultural machines present at that time, and the formula is expressed as:

[0105]

[0106] Where, E j (t) is the cooperative operation rate of farmland j at any time t, M j (t) represents the set of agricultural machines cooperating on farmland j at time t; e kdenotes the work rate of the kth agricultural machine;

[0107] The cooperative work completion time constraint stipulates that the completion time of the farmland task is determined by the integral relationship between the total demand and the cooperative work rate, and meanwhile, it is ensured that the task amount of each farmland needs to be completed, and the formula is expressed as:

[0108]

[0109] Wherein, Q j denotes the task amount of the farmland j;

[0110] The agricultural machine work duration constraint stipulates that the work duration of each agricultural machine at the farmland task point is obtained by the difference between the completion time of the farmland task and the actual arrival time of the agricultural machine, and the formula is expressed as:

[0111]

[0112] Wherein, τ jk is the work duration of the kth agricultural machine on the farmland j;

[0113] The variable domain limit constraint is used for describing that the decision variables of the scheduling path and the task allocation should satisfy the 0-1 binary requirement, and the formula is expressed as:

[0114]

[0115] It should be pointed out that: the real environment and influencing factors of the field work of the agricultural machine are complex and changeable, and in order to facilitate the research, the following assumptions are made for the model:

[0116] 1. The type and work capacity of the agricultural machine of the agricultural machine cooperative are the same, and the speed of the agricultural machine is constant during the transfer process;

[0117] 2. The agricultural machine failure is not considered in the present application;

[0118] 3. The farmland work point allows multiple agricultural machines to work cooperatively at the same time, and the total cooperative work rate is the linear superposition of the work rates of the agricultural machines;

[0119] 4. The agricultural machine needs to work continuously until the farmland work is completed to access the next farmland task;

[0120] 5. The agricultural machine returns to the agricultural machine cooperative after completing all the work tasks.

[0121] The characteristics of the model established in the present application are:

[0122] 1. Multiple agricultural machines are allowed to work in parallel at the same task point, and the cooperative work rate changes dynamically with the number of participating agricultural machines and the actual participation time (see the task coverage constraint and the cooperative work rate constraint).

[0123] 2. Introduce dynamic operation mechanism: first, the operation time of agricultural machinery is no longer preset as a constant, but is dynamically adjusted according to the actual arrival time and the operation state of the existing agricultural machinery, which reflects the dynamic characteristics of the operation time; second, the completion speed of the task is dynamically superimposed by the number of on-site agricultural machinery and its operation rate, showing a "cumulative effect" that changes over time, that is, the task completion rate is a variable related to the arrival time of agricultural machinery and the cooperative operation rate (see cooperative operation rate constraint, agricultural machinery operation time constraint and cooperative operation completion time constraint). This model realizes the dynamic coupling modeling of "cooperative operation rate-operation time-task completion time", which is closer to the actual field operation scene and has stronger practicability.

[0124] The characteristics of the application / differences from the traditional model are:

[0125] 1. The traditional model generally assumes that a piece of farmland is completed by a single agricultural machine independently, which cannot reflect the cooperative scene of "subsequent agricultural machine joining the unfinished task" in actual production.

[0126] 2. In the traditional MRTA or VRP framework, the task completion time is usually set as a static value, the operation duration does not change with the actual operation process, and the task completion is usually based on the completion of the last device, which lacks dynamic expression of the cooperative process, as shown in Table 1. As can be seen from Table 1, the current agricultural machinery scheduling modeling method generally does not consider the two key characteristics at the same time.

[0127] Table 1 Comparison of agricultural machinery scheduling models

[0128]

[0129] Step 2, in order to reduce the invalid moving distance of agricultural machinery in the field and improve the scheduling efficiency and convergence speed of the algorithm, the application fuses the spatial position of the field and the task load to construct a feature vector, so as to improve the accuracy and representativeness of clustering division. The silhouette coefficient is used as an evaluation index to adaptively determine the optimal cluster number, which can overcome the limitations of subjective setting of cluster number in traditional clustering methods. Based on the proportional allocation of agricultural machinery resources according to the total amount of tasks in the cluster, preliminary load balancing is realized, which is beneficial to the rationality of resource allocation, which includes the following steps:

[0130] Step 2-1, based on the spatial coordinates (x i ,y i ) of the field and the task quantity Q j of the field, a clustering feature vector is constructed; K-Means clustering algorithm is used to perform spatial clustering on all field points, and the cluster number k is iterated in the range of 2 to "total number of fields-1", and the optimal clustering division is selected by combining the silhouette coefficient, such as Figure 2The silhouette coefficient is used as the clustering effect evaluation index, which can measure the tightness within the cluster and the separation between clusters, and the larger the value, the better the clustering effect. The calculation formula is as shown below:

[0131]

[0132] Wherein, a(i) represents the average distance of sample i to other samples in the same cluster, and b(i) represents the average distance of sample i to all samples in the nearest neighbor cluster;

[0133] Step 2-2, according to the relative proportion of the total amount of tasks in each cluster, allocate the corresponding number of agricultural machinery resources to each cluster, and randomly allocate the agricultural machinery tasks of each agricultural machinery to the cluster to which it belongs, form an initial task allocation scheme, and construct an initial population of the algorithm, as shown in Figure 3 .

[0134] Step 3, as shown in Figure 4 , design a scheduling optimization algorithm based on a two-stage evolution mechanism.

[0135] The initial stage is to perform local search and crossover operation of agricultural machinery path within the cluster; the later stage is to perform agricultural machinery path reconstruction across clusters and trigger the task migration strategy based on load balancing; the optimization process in the initial stage helps the algorithm to explore high-quality scheduling schemes more efficiently under the premise of ensuring the stability of the cluster structure and the feasibility of the solution, which significantly improves the convergence speed of the solution. The later stage not only reconstructs the work task sequence of agricultural machinery between different clusters and strengthens the cross-regional collaborative work capability, but also introduces a task migration strategy based on load balancing to further optimize the load distribution of agricultural machinery while ensuring the feasibility of scheduling; the specific steps are as follows:

[0136] Step 3-1, in the initial stage, perform local search and crossover of individuals within the cluster for each cluster; use local search and local reordering to improve convergence speed, which includes the following steps:

[0137] Step 3-1-1, according to the cluster label in step 2-2, select agricultural machinery belonging to the same cluster;

[0138] Step 3-1-2, for the work path of agricultural machinery belonging to the same cluster, perform crossover operation using OX and PMX crossover operators, and perform local search using path inversion, path insertion, and path insertion three types of local search operators, and then optimize the work sequence within the cluster.

[0139] Step 3-2, in the later stage, perform cross-cluster crossover and variation based on the initial solution set, allowing cross-cluster work path reconstruction; at the same time, introduce a load balancing migration strategy to consider the possibility of cross-cluster collaborative work. It includes the following steps:

[0140] Step 3-2-1, remove the cluster label restriction, allow different cluster agricultural machines to cross and mutate across clusters;

[0141] Step 3-2-2, cross operation on the operation path of different genus cluster agricultural machines, and cross cluster cluster operation path reconstruction with three types of local search operators (see step 3-1-2);

[0142] Step 3-2-3, load balancing judgment based on the current population individuals: calculate the load balancing degree coefficient cv of each agricultural machine operation time: the ratio of the standard deviation and the average value of each agricultural machine working time, the calculation formula is as follows:

[0143]

[0144] Among them, std{workTime_List} and mean{workTime_List} represent the standard deviation and the average value of the working time of the agricultural machine; if cv < δ, the load is relatively balanced, and the number of migrated agricultural machines is 1 / 8 of the total number of agricultural machines; if cv > δ, it is considered that the load is relatively unbalanced, and the number of migrated agricultural machines is 1 / 4 of the total number of agricultural machines; δ is the load balancing coefficient;

[0145] Step 3-2-4, arrange the agricultural machines in ascending order of working time, select the low-load agricultural machines as the front migration agricultural machines, and the high-load agricultural machines as the rear migration agricultural machines;

[0146] Step 3-2-5, for each high-load agricultural machine, select the longest task in its task list as the migration candidate set;

[0147] Step 3-2-6, traverse the possible insertion position of the low-load agricultural machine to generate the migration candidate solution; select the optimal migration scheme through weighted comprehensive score; the migration score calculation formula is as follows:

[0148] Score = a · AR total + β · AT max + γ · Var

[0149]

[0150] Among them, σ orign , σ new respectively represent the total transfer time before and after migration, the maximum completion time before and after migration, and the agricultural machine load variance before and after migration; α, β, γ are weight coefficients, which can be set according to actual demand or decision preference.

[0151] Step 4, simulating the agricultural machinery cooperative operation process in real time through an event-driven decoding mechanism, and outputting a scheduling scheme; this step is used for target value calculation of each scheduling individual in the population and dynamic simulation of the scheduling process, and specifically includes the following steps:

[0152] Step 4-1, using a real number vector to encode the operation scheduling scheme of agricultural machinery, each decoding individual being a two-dimensional real number matrix, the row representing the agricultural machinery number and the column representing the farmland index accessed by the corresponding agricultural machinery, constituting the operation sequence of the agricultural machinery;

[0153] Step 4-2, event initialization and queue construction: according to the individual encoding result, the moving time of the agricultural machinery to the first farmland is calculated, an initial “arrival event” is generated and added to the event queue;

[0154] Step 4-3, arrival event processing: when a piece of agricultural machinery arrives at the target farmland, the system first updates the remaining operation amount of the task point, identifies the cooperative agricultural machinery set currently in the farmland, calculates the cooperative operation speed and the task completion time (the departure time of the cooperative operation agricultural machinery) according to the cooperative operation speed, and updates the agricultural machinery state to “in operation”; departure event processing: when the remaining operation amount of the farmland task point is zero, the task is considered to be completed; the actual operation time, task state and completion time of the cooperative agricultural machinery are updated, and the corresponding departure event is generated and added to the event sequence; at the same time, the agricultural machinery is scheduled to the next farmland task point in its task sequence, the arrival time is calculated, and the state is updated to “transferring”;

[0155] Step 4-4, loop decoding process: the decoding process continuously advances in an event-driven manner; the system selects the agricultural machinery with the earliest next event occurrence time by searching the scheduled departure time of the agricultural machinery currently in the “in operation” state and the scheduled arrival time of the agricultural machinery in the “transferring” state, and judges the event type according to the state: if the state is “in operation”, the “departure event” is triggered, and the task departure processing logic is executed; if the state is “transferring”, the “arrival event” is triggered, and the task arrival processing logic is executed; the decoding continues until all tasks are completed;

[0156] Step 4-5, outputting an event tuple list, the events being arranged in chronological order, and outputting the cumulative working time of each agricultural machinery and two target values; each event tuple includes the agricultural machinery number, the farmland number, the event time and the event type.

[0157] In order to illustrate the event triggering and operation progress dynamic updating process in the cooperative decoding mechanism, the present application selects a typical scenario for illustration:

[0158] Suppose there are two agricultural machines A and B, and the operation rate of each is 1 unit area / time unit. The sequence of agricultural machines A visiting farmland is F1→F2, and the agricultural machine B only visits F2. The task demand and path time of each farmland are set as follows: the task amount of farmland F1 is 10, and the task amount of F2 is 15; the moving time from the warehouse to F1 is 10, the moving time from the warehouse to F2 is 15, and the moving time from F1 to F2 is 5.

[0159] The decoding scheduling process is as follows:

[0160] The agricultural machine A arrives at F1 at t = 10, starts independent operation, and takes 10 time units, and completes the operation at t = 20 and sets off for F2 (will arrive at t = 25);

[0161] The agricultural machine B arrives at F2 at t = 15, independently operates until t = 25, and completes the task amount of 10; the agricultural machine A arrives at F2 at t = 25, joins the cooperative operation, cooperates with B to complete the remaining task amount of 5, and takes 2.5 time units; the two agricultural machines complete the operation and leave at t = 27.5, and the decoding event sequence is shown in Table 2:

[0162] Table 2 Decoding event sequence

[0163] Event number Farm machine number Farmland number Event time Event type E1 A F1 10 Arrival (1) E2 B F2 15 Arrival (1) E3 A F1 20 Departure (0) E4 A F2 25 Arrival (1) E5 A F2 27.5 Departure (0) E6 B F2 27.5 Departure (0)

[0164] Operation time length statistics:

[0165] Agricultural machine A: operation 10, cooperative operation 2.5, total 12.5;

[0166] Agricultural machine B: independent operation 10, cooperative operation 2.5, total 12.5;

[0167] The technical value of the decoding process lies in: through the dynamic triggering and processing of arrival and departure events, the real-time state of the agricultural machine at the task point is accurately tracked, and the operation rate, the remaining amount of the task, and the completion time of the task point are dynamically updated. The dynamic scheduling process in the cooperative operation of the agricultural machine is completely restored, and the dynamic modeling of the operation time length and the task completion time can be realized.

[0168] To verify the effectiveness of the model and algorithm, and cover different task density and coordination complexity scenarios, the present application selects six groups of data sets based on the real land modeling of Suixian County for performance evaluation. The data sets are divided into three categories according to the task size: small scale S class: |H|≤20, |F|≤5; medium scale M class: 20<|H|≤60; large scale L class: |H|>60, |F|>15. Among them, |H| represents the number of joint harvesters participating in scheduling, and |F| represents the farmland task points constructed based on the basic unit of towns in Suixian County. The threshold values of each scale are set based on the statistical data of wheat machine harvesting operations in Suixian County in the past three years. Model parameters: all agricultural machines are homogeneous joint harvesters, and the operation capacity and movement capacity are consistent. The movement rate is uniformly set to 0.48 km / min; the operation rate of the agricultural machine is 0.2 mu / min. That is, each agricultural machine can complete 0.2 mu of standard operation area per minute. The transfer time between farmlands is calculated by the ratio of the distance between farmlands and the movement rate. Algorithm parameters: population size 150, maximum iteration 300 generations, α, β, γ weight coefficients are all set to 1, δ load balancing coefficient is taken as 0.1, and each algorithm is independently run ten times. The comparison algorithms include: AC_ACO (ant colony algorithm), Memetic_ALG (memetic algorithm), NSGA_II algorithm and C2LEA algorithm.

[0169] Evaluation indicators include:

[0170] (1) IGD (inverse generational distance): measures the convergence of the solution set;

[0171] (2) HV (hyper volume): measures the diversity and distribution of the solution set.

[0172] The overall results of the four comparison methods are shown in Tables 3, 4 and Figure 5 , including IGD, HV index comparison on all data sets, Wilson rank sum test results with a significance level of 0.05, and total transfer cost comparison. "(+)", "≈", and "-" represent that the algorithm of the present application performs significantly better, statistically equivalent, and significantly worse than the comparison algorithm, respectively. Through the last row of data in Tables 4 and 5, it is shown that the number of test instances in which the algorithm of the present application performs significantly better, statistically equivalent, and significantly worse than the comparison algorithm.

[0173] Table 3 IGD index comparison of four algorithms on six data sets (the smaller the better)

[0174] Data set Ant colony algorithm Meme algorithm NSGA-Ⅱ algorithm Invention algorithm S_14_4 162.19±19.54 184.91±17.77 81.69±33.68 73.00±11.74 S_17_5 229.26±29.51 312.07±53.87 146.96±4.47 172.43±22.71 M_35_10 1634.25±187.51 1744.55±219.29 1242.40±101.57 87.10±16.80 M_52_15 2267.29±247.48 2216.58±247.39 3018.33±108.10 19.79±6.32 L_70_20 4646.96±219.06 5518.61±290.25 4460.42±210.35 46.27±15.77 L_106_30 8136.67±540.03 7705.37±553.46 6253.22±1238.54 835.05±13.26 (+) / (≈) / (-) 6 / 0 / 0 6 / 0 / 0 5 / 1 / 0

[0175] Table 4 HV index comparison of four algorithms on six data sets (the larger the better)

[0176]

[0177]

[0178] 1. Objective value performance analysis:

[0179] Firstly, from the IGD index, as shown in Table 3, based on the results of Wilson rank sum test, the algorithm of the application is significantly better than all the comparative algorithms in most cases, and is comparable to the NSGA-II algorithm only on the small-scale data set S_17_5. From the HV index, as shown in Table 4, the algorithm of the application has achieved significant advantages in all cases, which is mainly due to the cluster-guided initialization mechanism, two-stage evolution strategy and task migration mechanism based on load adopted by the algorithm, which realizes the rationalization of task division configuration in the evolution process, thereby effectively improving the global search ability.

[0180] Secondly, from the data set size dimension analysis: for small-scale data sets (named with S_), the algorithm of the application shows excellent performance in most scenarios, for example, in the S_14_4 data set, the algorithm of the application achieves the smallest IGD value and the largest HV value compared with the other three comparative algorithms; in the S_17_5 data set, although the IGD value is significantly improved compared with the ant colony algorithm and the meme algorithm, the IGD value is slightly increased compared with the NSGA-II algorithm. According to the analysis of the Wilson rank sum test results, there is no significant difference between the two. From the HV index, the algorithm of the application has obvious improvement compared with the NSGA-II algorithm, which shows that the algorithm of the application has better performance in the solution set diversity on the small-scale data set; although the convergence is comparable to the NSGA-II algorithm, the algorithm of the application has better performance in the solution set diversity. The main reason is that the number of small-scale task points is small, the clustering effect is not obvious, and the guidance for the subsequent optimization process is limited. For medium and large-scale data sets (named with M_ and L_ respectively), the algorithm of the application has achieved the optimal result among all the comparative algorithms. Especially in the L_70_20 data set, the algorithm of the application has a large improvement in both IGD and HV indexes, which shows that in the scene with significantly improved solution space complexity, the algorithm has stronger search efficiency and can obtain a solution set with better convergence and more uniform distribution. The reason is that the large-scale task point distribution is dense, the clustering effect is significant, which can provide a more explicit guidance path for the subsequent optimization; at the same time, the large task quantity also provides a wider scheduling space for the task migration strategy based on load, thereby fully exerting the synergistic effect of each sub-strategy.

[0181] 2. Actual cost comparison and scheduling process visualization analysis:

[0182] In order to more intuitively show the economic effect of each algorithm in actual scheduling, Figure 5The four algorithms are selected on each test data set on the Pareto front solution of the minimum completion time scheduling scheme, and the transfer cost is calculated accordingly. Among them, the transfer cost is obtained by multiplying the total transfer time in the scheme by the unit time cost (0.96r / min, corresponding to 2r / km). It should be pointed out that: since the working area of each farmland is fixed in the problem setting, the corresponding harvesting cost can be regarded as a fixed cost, so the variable cost mainly considered in the scheduling optimization is the transfer cost of the agricultural machine during the operation process, so the total transfer cost of the agricultural machine is essentially equivalent to the total operation cost in the scheduling process. From Figure 5 It can be observed that the algorithm proposed in the present application shows better performance in all data sets, compared with the ant colony algorithm, the meme algorithm and the NSGA-II algorithm, the average scheduling cost is reduced by about 25.6%, 27.1% and 27.0% respectively, which realizes significant cost saving, and verifies the practical application value of the algorithm in engineering scheduling. Further, Table 5 lists the detailed scheduling sequence of the minimum completion time scheme in the Pareto solution of the algorithm in the M_35_10 data set, including the working time distribution of each agricultural machine in the corresponding farmland task. Figure 6 Based on the scheme, the corresponding Gantt chart is drawn, which clearly presents the start and end time of each agricultural machine, the operation overlap and the task switching, and intuitively reflects the dynamic process of resource coordination and task execution.

[0183] Table 5 agricultural machine scheduling sequence and working time

[0184]

[0185] Note: The array in the square brackets in the table represents the task sequence of each agricultural machine, and the array element is the task point number executed by the agricultural machine in the order of operation. 3) Dynamic coordination mechanism visualization analysis:

[0186] In order to verify the actual running process of the dynamic multi-machine coordination mechanism in the constructed model, the present application selects some typical farmland task points to draw the "task accumulation-time" integral curve, as shown in Figure 7 By tracking the dynamic participation of each agricultural machine at different times, it can be observed that: with the change of the number of coordinated agricultural machines, the operation rate appears to be coordinated and superimposed at different stages, thereby affecting the accumulation rate of the task amount. The curve truly reflects the dynamic characteristics of the coordinated operation rate evolution with time, reveals the change law of the task advancing rhythm under the driving of the coordination mechanism, fully embodies the ability and dynamic adaptability of the model to depict the actual agricultural machine coordination operation process, and verifies the adaptability and effectiveness of the proposed method in the multi-agricultural machine dynamic scheduling environment.

[0187] The ablation experiment results are as follows:

[0188] To further verify the effectiveness of the proposed task migration strategy based on load balancing, the present application designs two control experiments, "task migration strategy based on load balancing" and "task migration strategy without load balancing", conducts ablation analysis to evaluate the influence of the strategy on algorithm convergence performance, solution set diversity and scheduling balance. The experiments are based on the aforementioned 6 typical data sets, and under the premise of keeping the rest of the algorithm parameters consistent, the "no strategy" control group is constructed by removing the task migration module. Figure 8 The ablation experiment results under two conditions are shown, and comparative analysis is carried out from the dimensions of IGD, HV and load variance.

[0189] The experimental results show that after introducing the task migration strategy based on load balancing, the algorithm of the present application has significant advantages in the three performance indicators. Among them, on large-scale data sets, the strategy has a particularly obvious effect on improving the quality of solutions. In summary, the introduction of the strategy not only significantly reduces the IGD value and improves the HV value, but also effectively balances the work load of each agricultural machine and reduces the overall scheduling fluctuation and resource vacancy of the system. It has strong practical application guiding significance.

[0190] Conclusion and application prospect:

[0191] 1. The present application proposes a dynamic cooperative mechanism agricultural machine scheduling modeling method and a two-stage multi-objective optimization algorithm integrating clustering guidance and load migration strategy. By constructing a "task volume-cooperative operation rate" integral model and an event-driven decoding mechanism, the applicability of traditional models in dynamic cooperative process modeling is improved. The two-stage C2LEA algorithm guided by clustering significantly improves the convergence speed and solution quality of the algorithm. At the same time, aiming at the problem of uneven load of agricultural machines, a dynamic task migration strategy based on load is proposed, which effectively improves the balance of task allocation and reduces the risk of failure caused by overload operation of agricultural machines, and has certain operation reliability and practical value.

[0192] 2. At the same time, experiments are carried out on the actual plot data set of Liuxian area, and the performances of C2LEA algorithm and mainstream multi-objective optimization algorithm in HV, IGD and total transfer cost indicators are compared. The introduction effect of the task migration strategy is evaluated through ablation analysis, especially its influence on the load variance of agricultural machines. In addition, by drawing the scheduling Gantt chart and the "farmland task volume-cooperative operation rate" integral curve, the dynamic evolution process of the algorithm scheduling process and the cooperative mechanism is intuitively displayed, further verifying the effectiveness and operability of the proposed model and algorithm.

[0193] 3. The present application provides a new modeling idea and optimization method for multi-agricultural machine cooperative scheduling in fragmented farmland areas, which is also applicable to intelligent operation management requirements in large field environments. Future research can be further expanded in the following directions:

[0194] (1) Agricultural machinery heterogeneity modeling, considering the performance of different types of agricultural machinery and the difference and collaborative adaptability of different types;

[0195] (2) Dynamic scheduling constraint enhancement, introducing time window, weather change, sudden mechanical failure and other uncertain factors;

[0196] (3) Combined with online rescheduling mechanism, realize intelligent agricultural machinery scheduling system with more real-time response ability.

Claims

1. A multi-objective optimization method for coordinated scheduling of multiple agricultural machines, characterized in that, Includes the following steps: Step 1, construct a system to minimize the maximum completion time for all farmland tasks: And minimize the total transfer time of all agricultural machinery: A multi-objective optimization model with the objective of [objective name missing]. Where N represents the amount of farmland; M represents the amount of agricultural machinery; t ij This represents the transfer time from farmland i to task point j; This indicates whether the k-th agricultural machine moves from farmland i to farmland j, with 1 indicating execution and 0 indicating no execution; ct i Indicates the completion time of farmland i; Step 2: A clustering algorithm is used to divide the farmland tasks into clusters based on their spatial distribution and workload. Then, the number of farm machines is dynamically allocated according to the cluster size. Subsequently, each farm machine is randomly assigned a farmland task within its own cluster to generate an initial task configuration, thereby constructing the initial population of the algorithm. Step 3: Optimize the scheduling scheme based on a two-stage evolutionary mechanism: Initial stage: Perform cross-path and local search operations for agricultural machinery within the cluster; Later stage: Perform agricultural machinery path reconstruction across clusters and trigger a load-balanced task migration strategy; Step 4: Accurately simulate the collaborative operation process of agricultural machinery through an event-driven decoding mechanism and output a scheduling scheme.

2. The multi-objective optimization method for multi-agricultural machinery collaborative scheduling according to claim 1, characterized in that, The multi-objective agricultural machinery scheduling optimization model satisfies the following constraints: The entry and exit balance constraint requires that each agricultural machine, within its scheduling path, must have clear "arrival" and "departure" operations for each plot of farmland it visits. The formula is expressed as: The task coverage constraint requires that the solution ensures that all farmland is fully operated by at least one agricultural machine, meaning that there are no uncovered task points in the system, and the formula is expressed as: The non-repetitive nature of the operation constraint means that the same agricultural machinery cannot perform multiple operations on the same field; the machinery can only be moved to the next task point after the workload on the target field has been reduced to zero. The formula is as follows: The scheduling starting point is normalized, and the initial state of all agricultural machinery is uniformly set to start from the scheduling center. Both the initial arrival time and start time are set to 0, serving as the unified time origin for the scheduling process. The formula is expressed as: Among them, at 0k This indicates the time when the k-th agricultural machine departs from the garage; ct0 indicates that the garage completion time is set to 0, meaning it does not need to be completed, and index 0 represents either the garage or the dispatch center; The time accessibility constraint stipulates that the time for each agricultural machine to reach the next task point should be equal to the sum of the completion time of its previous task point and the travel time between the two points. The formula is as follows: Among them, at jk This indicates the time when the k-th agricultural machine arrives at field j; The cooperative operation rate constraint states that the cooperative operation rate of farmland at any given time is the sum of the operation rates of the agricultural machinery present at that time, expressed by the formula: Among them, E j (t) represents the cooperative operation rate of farmland j at any time t, M j (t) represents the set of agricultural machines working together on farmland j at time t; e k This represents the operating speed of the k-th agricultural machine; The collaborative operation completion time constraint stipulates that the completion time of farmland tasks is determined by the integral relationship between the total task demand and the collaborative operation rate, while ensuring that the task volume of each farmland must be completed. The formula is expressed as: Among them, Q j Indicates the workload of farmland j; The agricultural machinery operation time constraint stipulates that the operating time of each agricultural machine at the farmland task point is calculated as the difference between the completion time of the farmland task and the actual arrival time of the agricultural machine, expressed by the formula: Where, τ jk Let k be the operating time of the kth agricultural machine on farmland j; Variable domain constraints, used to describe the decision variables for scheduling paths and task allocation, require that they satisfy the 0-1 binary requirement, expressed by the formula:

3. The multi-objective optimization method for coordinated scheduling of multiple agricultural machines according to claim 1, characterized in that, Step 2 specifically includes the following steps: Step 2-1, based on the spatial coordinates (x) of the farmland i ,y i ) and farmland task quantity Q j Cluster feature vectors are constructed; the K-Means clustering algorithm is used to spatially cluster all farmland points, with the number of clusters k ranging from 2 to "number of farmland points - 1". The optimal clustering is selected by combining the silhouette coefficient; the silhouette coefficient serves as an evaluation index for clustering performance, simultaneously measuring intra-cluster compactness and inter-cluster separation, with a larger value indicating better clustering performance; its calculation formula is shown below: Where a(i) represents the average distance from sample i to other samples in the same cluster, and b(i) represents the average distance from sample i to all samples in the nearest neighbor cluster; Step 2-2: Based on the relative proportion of the total number of tasks in each cluster, allocate a corresponding number of agricultural machinery resources to each cluster, and randomly assign agricultural machinery tasks to each agricultural machine belonging to its cluster to form an initial task configuration scheme and construct the initial population of the algorithm; wherein each individual in the population uses a real number vector encoding to represent the agricultural machine's operation scheduling scheme, and each individual is encoded as a two-dimensional real number matrix, with rows representing the agricultural machine number and columns representing the farmland index visited by the corresponding agricultural machine, which constitutes the operation sequence of the agricultural machine.

4. The multi-objective optimization method for multi-agricultural machinery collaborative scheduling according to claim 1, characterized in that, Step 3 specifically includes the following steps: Step 3-1: In the initial stage, the agricultural machinery set belonging to the same cluster is subjected to inter-individual crossover and local search operations using two crossover operators and three local search operators: path reversal, path exchange, path insertion, and path insertion. Step 3-2: In the later stage, based on the initial solution set, cross-cluster crossover and mutation are performed to allow cross-cluster job path reconstruction; at the same time, a task migration strategy based on load balancing is introduced to consider the possibility of cross-cluster collaborative work.

5. A multi-objective optimization method for coordinated scheduling of multiple agricultural machines according to claim 4, characterized in that: Step 3-1 specifically includes the following steps: Step 3-1-1: Based on the clustering labels in Step 2, select agricultural machinery belonging to the same cluster; Step 3-1-2: For the operation paths of agricultural machinery belonging to the same cluster, perform crossover operations using the OX and PMX crossover operators, and perform local search using three types of local search operators: path reversal, path insertion, and path insertion. Based on this, optimize the operation sequence within the cluster.

6. A multi-objective optimization method for coordinated scheduling of multiple agricultural machines according to claim 4, characterized in that: Step 3-2 specifically includes the following steps: Step 3-2-1: Remove cluster label restrictions, allowing cross-cluster crossover and variation of agricultural machinery from different clusters; Step 3-2-2: Perform cross-operation on the operation paths of agricultural machinery in different clusters using the OX and PMX cross-operation operators, and reconstruct the operation paths across clusters using three types of local search operators; Step 3-2-3, determine load balancing based on the current population: Calculate the load balancing coefficient cv for each agricultural machine's operating time: the ratio of the standard deviation to the average operating time of each agricultural machine, calculated using the following formula: Where std{workTime_List} and mean{workTime_List} represent the standard deviation and average value of the working time of agricultural machinery, respectively; if cv<δ, the load is relatively balanced, and the number of agricultural machinery to be moved is 1 / 8 of the total number of agricultural machinery; if cv>δ, the load is considered to be relatively unbalanced, and the number of agricultural machinery to be moved is 1 / 4 of the total number of agricultural machinery; δ is the load balancing coefficient. Step 3-2-4: Sort the agricultural machines in ascending order of operation time, select the agricultural machines that migrated first as low-load agricultural machines, and the agricultural machines that migrated last as high-load agricultural machines; Step 3-2-5: For each high-load agricultural machine, select the task with the longest time in its task list as the migration candidate set; Step 3-2-6: Traverse the possible insertion positions of the low-load agricultural machinery to generate candidate migration solutions; select the optimal migration scheme through weighted comprehensive scoring; the migration score calculation formula is as follows: Score=α·ΔR total +β·ΔT max +γ·ΔVar in, σ orign σ new These represent the total transfer time before and after the migration, the maximum completion time before and after the migration, and the variance of agricultural machinery load before and after the migration, respectively; α, β, and γ are weighting coefficients.

7. A multi-objective optimization method for coordinated scheduling of multiple agricultural machines according to claim 6, characterized in that: In the later stages of algorithm evolution, cross-cluster crossover and local search operations between individuals are allowed, thereby exploring potential global optimal solutions in a larger solution space. The load-balanced task migration strategy dynamically identifies high-load tasks and implements task migration based on multi-objective migration scores, thereby achieving resource redistribution at the task level.

8. A multi-objective optimization method for collaborative scheduling of multiple agricultural machines according to claim 1, characterized in that, Step 4 specifically includes the following steps: Step 4-1: The operation scheduling scheme of agricultural machinery is represented by real number vector encoding. Each decoded individual is a two-dimensional real number matrix, with the rows being the agricultural machinery number and the columns representing the farmland index visited by the corresponding agricultural machinery, which constitutes the operation sequence of the agricultural machinery. Step 4-2, Event Initialization and Queue Construction: Based on the individual coding results, calculate the movement time to the first farmland, generate an initial "arrival event" and add it to the event queue; Step 4-3, Arrival Event Handling: When a farm machine arrives at the target farmland, the system first updates the remaining workload at that task point, identifies the set of cooperating farm machines currently in that farmland, calculates the cooperative operation rate, and calculates the task completion time (the departure time of the cooperating farm machine) accordingly, and updates the farm machine status to "operating"; Departure Event Handling: When the remaining workload at the farmland task point is zero, the task is considered complete; the actual operation time, task status, and completion time of the cooperating farm machine are updated, and a corresponding departure event is generated and added to the event sequence; at the same time, the farm machine is scheduled to the next farmland task point in its task sequence, its arrival time is calculated, and its status is updated to "transferring"; Step 4-4, Loop Decoding Process: The decoding process proceeds continuously in an event-driven manner. The system retrieves the scheduled departure time of agricultural machinery currently in the "Working" state and the scheduled arrival time of agricultural machinery in the "Transferring" state, selects the agricultural machinery with the earliest next event occurrence time, and determines the event type based on its state: if the state is "Working," a "Departure Event" is triggered, and the task departure processing logic is executed; if the state is "Transferring," an "Arrival Event" is triggered, and the task arrival processing logic is executed; decoding continues until all tasks are completed. Steps 4-5: Output a list of event tuples, with events arranged in chronological order. Output the cumulative working time and two target values ​​for each agricultural machine. Each event tuple includes the agricultural machine number, farmland number, event time, and event type.

Images