Multi-target agricultural machinery scheduling method for cross-regional mountain farmland management

Abstract

The invention discloses a multi-objective optimization method for cross-regional scheduling of agricultural machinery in multiple regions of a mountainous region, and the method comprises the steps: firstly, employing an improved multi-constraint density peak clustering algorithm, calculating the local density through a Gaussian kernel function, and carrying out the intelligent grouping of scattered farmlands through comprehensive consideration of multiple factors, such as geographical distance, terrain constraint, time window limitation, and the like; secondly, establishing a dual-objective optimization model covering starting cost, distance cost, penalty cost, waiting time, transfer time and operation time, and meanwhile, minimizing scheduling cost and completion time; a core adopts a memory enhanced grey wolf optimization (MEGWO) algorithm to solve, integrates improvement strategies such as optimal point set initialization, a double-population memory mechanism, a differential evolution operator, random local search and linear population reduction, and processes a multi-objective optimization problem in combination with non-dominated sorting and a crowding distance mechanism. And the intelligent level and the operation efficiency of agricultural machinery dispatching under the complex mountain terrain are obviously improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of agricultural machinery scheduling optimization, in particular to a multi-objective optimization method for multi-region agricultural machinery cross-region scheduling in mountainous areas. BACKGROUND

[0002] With the rapid development of modern agriculture and the continuous improvement of agricultural mechanization, agricultural machinery cross-region scheduling has become an important means to improve agricultural production efficiency and reduce operating costs. In particular, in the vast mountainous and hilly areas, the distribution of farmland is scattered, the terrain is complex, and the traffic conditions are limited, so the traditional agricultural machinery scheduling method cannot meet the actual demand.

[0003] The existing agricultural machinery scheduling technology mainly has the following problems:

[0004] Firstly, the traditional scheduling method mainly focuses on cost minimization or time minimization, and lacks consideration of multi-objective collaborative optimization. In actual application, agricultural production not only requires cost control, but also needs to ensure the timeliness of operation, especially in the critical agricultural season, the time window constraint becomes an important factor.

[0005] Secondly, the existing technology has a sharp increase in computational complexity when dealing with large-scale, cross-region scheduling problems, and the optimization efficiency is low. Traditional exact algorithms such as integer programming often cannot find the optimal solution within a reasonable time when facing complex scenarios of multiple machines and multiple farmlands, while heuristic algorithms are prone to local optimum. At the same time, the complexity of the terrain and the dispersion of farmland in mountainous and hilly areas make clustering grouping a key pre-step for scheduling optimization, but existing clustering methods are mostly based on simple distance or area indicators, lacking comprehensive consideration of multi-dimensional factors such as terrain constraints and time window restrictions.

[0006] In recent years, swarm intelligence optimization algorithms have shown good performance in combinatorial optimization problems, among which the grey wolf optimization algorithm (GWO) has attracted attention due to its simple structure and fewer parameters. However, the traditional GWO algorithm has problems such as insufficient population diversity, easy premature convergence, and unbalanced exploration and exploitation capabilities, which limit its effectiveness in dealing with complex optimization problems with multiple objectives and multiple constraints. SUMMARY

[0007] The purpose of the present application is to solve at least one of the above technical defects.

[0008] To this end, one object of the present application is to propose a multi-objective agricultural machinery scheduling method for cross-region mountainous farmland management to solve the problems mentioned in the background art and overcome the deficiencies in the prior art.

[0009] In order to achieve the above-mentioned purpose, the present application provides a multi-objective agricultural machinery scheduling method for cross-region mountainous farmland management, comprising:

[0010] Multi-modal data collection and pre-processing, the data including geographic coordinates, area information, job requirements and time window constraints;

[0011] Multi-constrained density peak clustering grouping, including applying a multi-constrained density peak clustering algorithm to divide the farmland plots and assign them as geographic clusters;

[0012] Dual-objective optimization model construction, including establishing a dual-objective optimization model to minimize the total scheduling cost and the total completion time, wherein the scheduling cost includes start-stop cost, travel distance cost and penalty cost, and the completion time includes waiting time, transfer time and job time;

[0013] Memory-enhanced grey wolf optimization solution, including initializing the population of the memory-enhanced grey wolf optimization algorithm using an initialization method based on the best point set to generate an initial solution;

[0014] Non-dominated sorting and crowded distance selection, including using a non-dominated sorting and crowded distance mechanism to handle multiple objective functions;

[0015] Grey wolf position updating and leader selection mechanism, including updating the position of the wolf pack under the guidance of alpha wolf, beta wolf and delta wolf through an enhanced position updating formula, and calculating the fitness value of each individual;

[0016] The memory mechanism is realized by maintaining the exploration group and the memory group, wherein the good solutions are stored and the bad solutions are eliminated;

[0017] Evolution operator integration, including introducing mutation, crossover and selection operators based on the differential evolution strategy to improve search efficiency;

[0018] Random local search and linear population reduction, including selecting some wolves in the memory group to perform random local search;

[0019] Linear population size reduction strategy is adopted, and the population size is dynamically reduced according to the fitness evaluation;

[0020] Dynamic adjustment and real-time optimization of scheduling scheme, including generating the optimal scheduling path and resource allocation scheme, which clearly defines the allocation of each agricultural machine, the operation order and the operation time of each farmland.

[0021] As preferred, the multi-constrained density peak clustering algorithm includes:

[0022] Calculate the density value of each farmland plot based on geographic coordinates and job constraints;

[0023] Determine the core point, boundary point and noise point according to the field radius and minimum point number parameters;

[0024] Identify the cluster center in the high-density area;

[0025] Expanding the cluster around the cluster center to group the farmland plots that satisfy the spatial and temporal constraints.

[0026] As preferred, the optimal point set initialization comprises:

[0027] Generating a canonical point set in s-dimensional Euclidean space by multiplying the basis function value of each dimension with the corresponding ordinal number to construct the point set coordinates;

[0028] Calculating a deviation value to evaluate the distribution quality of the point set, the deviation value being determined by a specific mathematical relationship involving constants related to the geometric parameter g and a positive error parameter ε;

[0029] Setting the geometric parameter g as a cosine function sequence, the cosine function sequence being constructed based on the dimension ordinal number and a specific prime number p, wherein the dimension ordinal number ranges from 1 to s, and the prime number p satisfies the condition that it is not less than the dimension number s after subtracting 1.5, and the minimum prime number value satisfying this condition is selected.

[0030] As preferred, the memory mechanism comprises:

[0031] Maintaining an exploration population and a memory population, and the exploration population and the memory population have equal sizes;

[0032] After initialization, the wolf with the best performance is stored in the memory population;

[0033] Updating the exploration population position in each iteration process, and refreshing the memory population with improved solutions;

[0034] Comparing the cost function value of the memory population with the cost function value of the exploration population leader wolf, and replacing the leader wolf when the memory population has a better solution.

[0035] As preferred, the evolution operator comprises:

[0036] Applying a mutation operation, taking the position of the current optimal wolf α as the benchmark, combining a dynamic scaling factor and the difference between the optimal position and the random wolf position to generate a mutation vector;

[0037] Dynamically adjusting the scaling factor, which adopts a larger value in the early stage of the algorithm to enhance the global exploration ability, and linearly decreases to a smaller value along with the iteration process;

[0038] Performing a crossover operation on the target vector and the mutation vector according to a preset crossover probability, and randomly selecting vector components to form trial individuals;

[0039] Using a greedy selection strategy to compare the fitness of the trial vector and the target vector, and preferentially retaining individuals with better fitness as the next generation population members.

[0040] As preferred, the random local search comprises:

[0041] Select the top half of individuals with the best performance from the memory population as the local search object according to the fitness ranking, and balance the search efficiency and the calculation cost;

[0042] Calculate the Euclidean distance between each selected individual and other individuals in the solution space, and identify the nearest neighbor individual;

[0043] According to the fitness of the neighbor individual, different position generation strategies are adopted, when the neighbor individual performs better, the new position is generated by approaching it through interpolation method, and when the neighbor individual performs worse, the new position is generated by moving away from it through extrapolation method;

[0044] The fitness of the generated new position is evaluated, and only when the fitness of the new position is better than the current position, the replacement is performed, otherwise the original position remains unchanged.

[0045] As preferred, the linear population size reduction comprises:

[0046] A larger initial population size is set at the beginning of the algorithm, and a smaller minimum population size is set at the end of the algorithm;

[0047] According to the ratio of the current function evaluation times to the maximum allowed evaluation times, the target population size of the next generation is linearly calculated;

[0048] After each iteration is completed, the exploration population and the memory population are respectively arranged in descending order according to the fitness, and the worst individual is identified;

[0049] According to the target population size calculated, the individual with the lowest fitness is eliminated from the end of the two types of populations, and the population size is gradually reduced to retain high-quality individuals.

[0050] Compared with the prior art, the present application has the advantages and beneficial effects that:

[0051] The multi-objective agricultural machinery scheduling method for cross-regional mountain field management provided by the present application adopts a multi-objective collaborative optimization method, simultaneously considers cost and time, and provides multiple balanced schemes for users; the multi-constraint density peak clustering algorithm is used to realize intelligent grouping of fields, so as to reduce the problem size and the calculation complexity; the improved MEGWO algorithm is introduced, so as to enhance the optimization performance and overcome the premature convergence and local optimum problems of traditional methods; and the overall effect is that the scheduling cost is reduced, the operation time is shortened, the mechanical demand is reduced, the system response is faster, and the mechanical utilization rate is improved.

[0052] Additional aspects and advantages of the application will be set forth in part in the description which follows, and in part will become apparent to those skilled in the art upon examination of the following and / or can be learned by practice of the application. BRIEF DESCRIPTION OF DRAWINGS

[0053] The above and / or additional aspects and advantages of the present application will become apparent and more readily appreciated from the following description, taken in conjunction with the following drawings of which:

[0054] Figure 1 is a whole flow chart of a multi-target agricultural machinery scheduling method for cross-regional mountainous farmland management provided by an embodiment of the present application.

[0055] Figure 2 is a multi-target agricultural machinery cross-regional scheduling schematic diagram provided by an embodiment of the present application.

[0056] Figure 3 is a detailed flow chart of a multi-constraint density peak clustering algorithm in an embodiment of the present application.

[0057] Figure 4 is a user interface design and interaction flow chart in an embodiment of the present application.

[0058] Figure 5 is a performance comparison convergence curve diagram in an embodiment of the present application.

[0059] Figure 6 is a route planning and optimization result visualization diagram in an embodiment of the present application. DETAILED DESCRIPTION

[0060] Embodiments of the present application are described in detail below, examples of which are shown in the accompanying drawings, wherein the same or similar reference numerals represent the same or similar elements or elements having the same or similar functions throughout. The embodiments described below by referring to the accompanying drawings are exemplary and are intended to explain the present application, and cannot be understood as a limitation of the present application.

[0061] Embodiment 1:

[0062] The technical problem to be solved by the present application is to overcome the single-target optimization limitations, low computational efficiency, poor dynamic adaptability and other problems in the mountainous multi-regional agricultural machinery scheduling algorithm of the prior art, and to provide a multi-target optimization method considering cost and time comprehensively, to realize efficient and intelligent cross-regional agricultural machinery scheduling.

[0063] To achieve the above-mentioned purpose, one aspect of the present application provides a multi-target optimization method for mountainous multi-regional agricultural machinery cross-regional scheduling. The method adopts a two-stage optimization strategy of "clustering first, scheduling later", first intelligently groups the scattered farmland through an improved multi-constraint density peak clustering algorithm, and then solves the multi-target scheduling model by using a memory-enhanced grey wolf optimization algorithm.

[0064] The embodiment of the application aims at the practical problems of low efficiency, high cost and poor adaptability of agricultural machinery scheduling under complex mountainous terrain conditions, and proposes a complete multi-objective optimization solution. The method takes the actual agricultural machinery scheduling demand of Dechang County in Liangshan Prefecture of Sichuan Province as the application background, and realizes efficient allocation of agricultural machinery resources through systematic technological innovation.

[0065] As shown in Figures 1-6 , a multi-objective agricultural machinery scheduling method for cross-regional mountainous farmland management includes:

[0066] Step S1: multi-modal data acquisition and preprocessing, collecting farmland data from multiple agricultural cooperatives, including geographic coordinates, area information, operation demand and time window constraints.

[0067] This step comprehensively collects multi-dimensional data such as farmland geographic coordinates, terrain features, area information, operation demand, and time window constraints, and establishes a complete problem data set. The GPS positioning technology is used to obtain geographic coordinates with an accuracy of ±1 meter, the digital elevation model (DEM) is used to extract terrain parameters such as slope and elevation difference, and meteorological data and crop growth information are used to determine reasonable operation time windows. Data preprocessing includes outlier detection, data standardization and missing value completion to ensure the quality of data for subsequent analysis.

[0068] Specifically, first, the accurate boundary coordinates of the farmland plot are collected by high-precision GPS positioning equipment, with positioning accuracy controlled within ±1 meter, and the WGS84 coordinate system is used to ensure data consistency. For areas with weak GPS signals in the mountains, the system is supplemented with differential GPS or RTK technology to improve positioning accuracy. Terrain feature data is obtained through digital elevation model (DEM), which extracts key parameters such as slope, aspect, and elevation difference of each farmland plot. The DEM data resolution reaches 5m x 5m, providing accurate basis for mechanical trafficability assessment and operation difficulty analysis. The recording of farmland attribute information covers area (accurate to 0.01 hectares), crop type, soil condition, irrigation facilities and other elements, and the operation demand is calculated by multiplying the farmland area by the unit area operation quantity. Detailed registration of mechanical equipment specifications includes mechanical model, loading capacity, maximum operation distance, operation speed, fuel consumption rate and other parameters, providing accurate constraint conditions for subsequent optimization calculation.

[0069] At the same time, all schedulable agricultural machinery is registered for specifications, covering loading capacity Q m , maximum operation distance D m , operation speed V m and other key parameters, and setting corresponding operation time windows [B i , E iOn this basis, the collected data is preprocessed, including outlier detection, missing value completion and standardization processing, to ensure the accuracy and availability of the data

[0070] Step S2: Multi-constraint density peak clustering grouping, including applying a multi-constraint density peak clustering algorithm to divide the farmland plots and assign them to geographical clusters, wherein each cluster is assigned to a specific agricultural cooperative based on spatial proximity and job constraint conditions.

[0071] Further, the multi-constraint density peak clustering algorithm includes:

[0072] Calculating the density value of each farmland plot based on geographical coordinates and job constraints;

[0073] Determining core points, boundary points and noise points according to the field radius and minimum point number parameters;

[0074] Identifying cluster centers in high-density areas;

[0075] Expanding clusters around cluster centers to group farmland plots that meet spatial and temporal constraints.

[0076] An improved density peak clustering algorithm is proposed for the complexity of mountainous farmland distribution. The algorithm first calculates the local density of each farmland plot and uses a Gaussian kernel function for density estimation. Then, the relative distance is calculated and the plots with high density and large relative distance are identified as cluster centers. Based on multiple constraint conditions such as terrain constraints, distance limits and job capacity, the farmland plots are divided into several geographical clusters, each cluster is assigned to the most suitable agricultural cooperative, so as to divide the global multi-cooperative scheduling problem into multiple local single-cooperative scheduling sub-problems.

[0077] Specifically, an improved density peak clustering method is used to calculate the local density of farmland plots by a Gaussian kernel function where d ij is the geographical distance between farmland plots i and j, d c is the cutoff distance parameter. The system also calculates the minimum relative distance δi = min{d ij : ρj > ρi} of each plot to a higher density plot, and identifies the plots with high local density and large relative distance as cluster centers. Based on traditional clustering, the system integrates multiple constraint conditions such as terrain constraints (slope difference control), distance constraints (maximum distance limit within the cluster), capacity constraints (total demand and service capacity matching), and time constraints (time window overlap requirement), to ensure that the grouping result not only conforms to the geographical distribution law, but also meets the actual job requirements.

[0078] Step S3: dual-objective optimization model construction, including establishing a dual-objective optimization model while minimizing total scheduling cost and total completion time, wherein the scheduling cost includes start-stop cost, driving distance cost and penalty cost, and the completion time includes waiting time, transfer time and working time.

[0079] Meanwhile, considering the two core objectives of cost minimization and time optimization, a dual-objective optimization model is established which comprehensively considers scheduling cost and completion time. Objective function 1 is to minimize total scheduling cost min C =C start +C distance +C penalty , including mechanical start-up cost C start , driving distance cost C distance and time window violation penalty cost C penalty . Objective function 2 is to minimize total completion time min T =T wait +T transfer +T operate , including waiting time T wait , transfer time T transfer and working time T operate . Meanwhile, multiple constraint conditions such as mechanical capacity constraint, distance limit constraint, demand satisfaction constraint and time window constraint are set to ensure the feasibility and practicality of the scheme.

[0080] Specifically, the constraint condition setting includes flow balance constraint to ensure that each machine starts from the cooperative and returns:

[0081]

[0082] wherein x ijm is a binary decision variable, 1 indicating that machine m goes from location i to farmland j, and 0 indicating otherwise; A is the number of agricultural machinery cooperatives; F is the total number of farmland plots; M is the total number of schedulable agricultural machines; i is the starting location index; j is the target farmland index; and m is the agricultural machine index.

[0083] Service constraint to ensure that each farmland is only served by one machine:

[0084]

[0085] wherein f is the farmland plot set.

[0086] Capacity constraint, total driving distance does not exceed the working distance of the machine:

[0087]

[0088] wherein D m is the maximum working distance limit of machine m, unit: kilometers (km); dij is the actual distance from location i to location j, unit: kilometers (km); F+A is the total number of locations (number of farmland + number of cooperatives).

[0089] Demand constraint, total demand does not exceed total service capacity:

[0090]

[0091] where Q m is the working capacity of machine m, unit: cubic meters (m 3 ); q i is the working demand of farmland i, unit: cubic meters (m 3 ).

[0092] Time window constraint, working time must be within the allowed time window:

[0093]

[0094] P m is the time window violation penalty cost of machine m, unit: yuan; Bi is the earliest working start time of farmland i, unit: hours; E i is the latest working end time of farmland i, unit: hours; T mi is the actual time of machine m arriving at farmland i, unit: hours; C w is the early arrival penalty coefficient, unit: yuan / hour; C l is the late arrival penalty coefficient, unit: yuan / hour.

[0095] The model establishes a complete constraint system, including flow balance constraint to ensure that each machine departs from the cooperative and returns, service constraint to ensure that each farmland is only served by one machine, capacity constraint to limit the total travel distance to be less than the working radius of the machine, demand constraint to ensure that the total service capacity meets the working demand, and time window constraint to ensure that the working is completed within the allowed time period.

[0096] Step S4: memory-enhanced grey wolf optimization solution, including initializing the population of the memory-enhanced grey wolf optimization (MEGWO) algorithm using a best point set (BPS)-based initialization method to generate initial solutions with reasonable distribution.

[0097] Further, the best point set initialization includes:

[0098] Generating a standard point set in an s-dimensional Euclidean space, constructing point set coordinates by multiplying the basis function value of each dimension with the corresponding serial number;

[0099] Calculating a deviation value to evaluate the distribution quality of the point set, the deviation value being determined by a specific mathematical relationship involving constants related to the geometric parameter g and the positive error parameter ε.

[0100] The geometry parameter g is set as a cosine function sequence, the cosine function sequence is constructed based on a dimension sequence and a specific prime number p, wherein the dimension sequence ranges from 1 to s, the prime number p satisfies the condition that it is not less than the dimension number s after subtracting 1.5, and the minimum prime value satisfying the condition is selected.

[0101] In view of the shortcomings of the traditional GWO algorithm, a memory-enhanced grey wolf optimization (MEGWO) algorithm is proposed. The algorithm integrates multiple improved strategies such as best point set initialization, memory mechanism, evolution operator, random local search and linear population reduction. First, the best point set theory is used to generate an initial population with uniform distribution, improving the global search ability of the algorithm. Then, two parallel populations, the exploration population and the memory population, are maintained. The former is responsible for global exploration, and the latter stores elite solutions and performs local development. The evolution operator is based on the differential evolution strategy, which enhances the adaptability of the algorithm by dynamically adjusting the mutation and crossover parameters. The random local search performs neighborhood search on the excellent individuals in the memory population, improving the quality of the solutions. The linear population reduction technique dynamically adjusts the population size, balancing the calculation efficiency and optimization effect.

[0102] Specifically, the algorithm uses the best point set theory for population initialization, and uses the The standard point set is generated, improving the global search ability.

[0103] P n (k) is the coordinate of the kth point in s-dimensional space; is the geometric parameter of the ith dimension, based on a cosine function sequence;

[0104] n is the total number of points in the point set; s is the dimension number of the problem space; k is the serial number of the point, k∈{1,2,...,n};

[0105] The population size is dynamically adjusted between 50-200 according to the problem complexity, and the coding scheme uses real value coding to map the scheduling decision to a floating point vector in the [0,1] interval. The decoding strategy determines the farmland access priority through vector sorting, and generates a specific scheduling plan combined with the mechanical capacity and time window constraints.

[0106] Step S5: non-dominated sorting and crowding distance selection, including using a non-dominated sorting and crowding distance mechanism to handle multiple objective functions. The non-dominated sorting and crowding distance mechanism is used to handle multiple objective functions and maintain population diversity.

[0107] To handle multi-objective optimization problems, the non-dominated sorting mechanism is used to stratify solutions, and the crowding distance is calculated to maintain the diversity of solutions. Through Pareto frontier analysis, a series of optimization schemes balancing cost and time are provided for decision makers.

[0108] Meanwhile, the learning ability and convergence performance of the algorithm are enhanced. The system establishes a framework for the parallel operation of the exploration group and the memory group. The exploration group is responsible for global search and maintains the algorithm's exploration ability. The memory group stores the best solutions of previous generations and continuously updates them through the elite selection strategy. Information exchange between the two populations ensures the preservation and spread of excellent solutions, and the elite protection strategy ensures that the best solution of each generation is not lost.

[0109] A double-population memory mechanism is designed to establish a double-population framework for the parallel operation of the exploration group and the memory group. The exploration group is responsible for global search and updates its position according to the standard GWO mechanism. The population size is the same as the initial population, and its main role is to maintain the algorithm's exploration ability. The memory group stores the best solutions of previous generations, and the population size is equal to that of the exploration group. The exploration group is copied at the initialization stage, and the subsequent updates are made through the elite selection strategy. Information exchange is performed at the end of each generation by comparing the solution quality of the two populations. The excellent solutions in the exploration group are copied to the memory group, replacing the inferior solutions in the memory group. Elite protection ensures that the best solution of each generation is not lost, and the solution quality is maintained through the elite protection strategy.

[0110] Step S6: Grey wolf position update and leadership selection mechanism, including updating the position of the wolf group under the guidance of alpha, beta, and delta wolves through the enhanced position update formula, and calculating the fitness value of each individual;

[0111] The memory mechanism is achieved by maintaining the exploration group and the memory group (population size equal). Excellent solutions are stored, and poor solutions are eliminated.

[0112] Further, the memory mechanism comprises:

[0113] The exploration group and the memory group are maintained, and the size of the exploration group and the memory group is equal;

[0114] After initialization, the best-performing wolf is stored in the memory group;

[0115] The exploration group position is updated in each iteration, and the memory group is refreshed with improved solutions;

[0116] The cost function values of the memory group and the exploration group leader wolves are compared, and the leader wolves are replaced when the memory group has better solutions.

[0117] The grey wolf position update and leadership selection mechanism achieves efficient solution space search. The algorithm selects the three wolves with the best fitness from the current population as alpha, beta, and delta leaders to guide the movement direction of other individuals. The position update adopts a multi-leader fusion strategy, which calculates the new position through the position update formula to achieve a smooth transition from global search to local search. The position update formula is:

[0118]

[0119] X1 j X2 j X3 j are the jth component of the candidate position vector based on three leaders respectively; X α,j X β,j X δ,j are the position component of the jth dimension of alpha wolf, beta wolf and delta wolf respectively; X i,j is the position component of the jth dimension of the current wolf i; θ is the convergence factor, ranging from [0, 2]; rand(0, 1) is a random number in the interval [0, 1]; j represents the dimension index, j ∈ {1, 2,..., D}, and D is the problem dimension.

[0120] Step S7: Evolution operator integration, including introducing mutation, crossover and selection operators based on the differential evolution strategy to improve search efficiency.

[0121] Further, the evolution operator includes:

[0122] A mutation operation is applied, taking the position of the current optimal wolf alpha as the benchmark, combining a dynamic scaling factor and the difference between the optimal position and the random wolf position to generate a mutation vector;

[0123] The scaling factor is dynamically adjusted, which is larger at the beginning of the algorithm to enhance global exploration ability, and linearly decreases to a smaller value with the iteration process;

[0124] According to the preset crossover probability, the target vector and the mutation vector are subjected to a crossover operation, and a vector component is randomly selected to form a trial individual;

[0125] A greedy selection strategy is used to compare the fitness of the trial vector and the target vector, and the individual with better fitness is preferentially retained as the next generation population member.

[0126] It can be understood that the larger value and the smaller value can be determined by manual setting or setting threshold.

[0127] This step further improves the algorithm performance, and the system introduces the mutation operation of differential evolution:

[0128] wherein is the jth component of the t+1th generation mutation vector, X j is the jth position component of the current individual, X alpha is the current optimal solution, F is the dynamic scaling factor, and t is the current iteration number. The dynamic scaling factor f min is the minimum scaling factor, f max is the maximum scaling factor, Maxit is the maximum iteration number, and it is the current iteration number.

[0129] The crossover operation targets the vector with the mutation vector for crossover:

[0130]

[0131] The selection operation adopts a greedy selection strategy:

[0132]

[0133] wherein is the jth dimension component of the (t+1)th generation trial vector; is the jth dimension component of the mutation vector; is the jth dimension component of the tth generation target vector; is the objective function value of the current target vector; Pc is the crossover probability, usually set to 0.7-0.9; rand(j) is a random number in the jth dimension, ranging from 0 to 1; and D represents the number of problem dimensions.

[0134] The crossover operation adopts a probability control strategy, balancing the diversity and convergence speed of the search. The selection operation follows the greedy principle, retaining only the individuals with better fitness.

[0135] Step S8: Random local search and linear population reduction are performed, including selecting part of the wolves in the memory population to perform random local search to improve the refinement ability of the solution;

[0136] A linear population size reduction (LPSR) strategy is adopted to dynamically reduce the population size according to the fitness evaluation.

[0137] Further, the random local search includes:

[0138] The best half of the individuals in the memory population are selected as the local search objects according to the fitness ranking, balancing the search efficiency and computational cost;

[0139] The Euclidean distance between each selected individual and other individuals in the solution space is calculated to identify the nearest neighbor individual;

[0140] Different position generation strategies are adopted according to the fitness of the neighbor individual, generating a new position close to the neighbor individual through interpolation when the neighbor individual performs better, and generating a new position away from the neighbor individual through extrapolation when the neighbor individual performs worse;

[0141] The fitness of the generated new position is evaluated, and the current position is replaced only when the fitness of the new position is better than that of the current position, otherwise the original position remains unchanged.

[0142] Specifically, the local search is performed on the elite individuals in the memory population. The search objects are selected from the top 50% of the individuals in the memory population in terms of fitness, balancing the computational efficiency and search quality. The neighborhood construction finds the nearest neighbors of the selected individuals based on the Euclidean distance, constructing the local search neighborhood. The position generation generates new positions based on the fitness of the neighbors:

[0143] X t,j = X i,j + C1 x rand(0, 1) x (X n,j - X i,j ) (when the neighbor is better)

[0144] X t,j = X i,j + C1 x rand(0, 1) x (X i,j - X n,j ) (when the neighbor is worse),

[0145] where X t,j is the j-th dimension component of the trial position; X i,j is the j-th dimension position component of the current individual i; X n,j is the j-th dimension position component of the nearest neighbor individual; rand(0, 1) represents a random number in the interval [0, 1]. The acceptance criterion accepts the new position only when it is better, otherwise the original position is maintained.

[0146] Further, the linear population size reduction comprises:

[0147] setting a larger initial population size at the beginning of the algorithm and a smaller minimum population size at the end of the algorithm;

[0148] linearly calculating the target population size of the next generation according to the ratio of the current number of function evaluations to the maximum allowed number of evaluations;

[0149] arranging the exploration population and the memory population in descending order of fitness after each iteration is completed, respectively, and identifying the worst-performing individuals;

[0150] eliminating the individuals with the lowest fitness from the end of the two populations according to the target population size calculated, and gradually reducing the population size to retain high-quality individuals.

[0151] It can be understood that the larger initial population size and the smaller minimum population size can be determined by manual setting or setting threshold values.

[0152] This step optimizes the computational efficiency, and the system performs a local fine search on the top 50% of the individuals in the memory population in terms of fitness, further improving the solution quality through neighborhood construction and position generation strategies. The linear population reduction mechanism adopts Dynamic adjustment of population size significantly reduces the computational overhead while maintaining the solution quality,

[0153] where N G+1 is the target population size of the G+1 generation; P min is the minimum population size, usually set to 10; P int is the initial population size, usually set to 100-200; MaxNFEs is the maximum number of function evaluations; NFE represents the current number of function evaluations; round is the rounding function; G represents the current generation number

[0154] Step S9: Dynamic adjustment and real-time optimization of the scheduling scheme, including generating the optimal scheduling path and resource allocation scheme, which clearly allocates each agricultural machine, the operation order, and the operation time of each farmland.

[0155] The present application realizes effective balance of cost and time. The system stratifies the population according to the non-dominated relationship, calculates the crowding distance D(i) = ∑ j＝1 D(i,j) m Maintains the diversity of solutions and provides multiple Pareto optimal solutions for decision makers.

[0156] where D(i) is the total crowding distance of individual i; D(i,j) is the crowding distance of individual i on the jth objective; m is the number of objective functions (m = 2 in this patent, including cost and time objectives); j is the objective function index.

[0157] The convergence criterion adopts multiple termination conditions, including the maximum number of iterations, continuous generation without improvement, objective function change threshold, etc., to ensure that the algorithm converges to high-quality solutions within a reasonable time.

[0158] In the actual verification in Dechang County, the optimization method shortens the spring fertilization operation, which originally takes 15 days, to 9 days, reduces the total cost from 42,000 yuan to 32,500 yuan, and improves the mechanical utilization rate from 65% to 88%. The multiple alternative schemes generated by the system enable the cooperative society to flexibly choose according to the financial situation and time requirements, significantly improving the scientific and fine level of mountainous agricultural machine scheduling.

[0159] Example 2:

[0160] The present application provides a multi-objective optimization system for mountainous multi-region agricultural machine cross-region scheduling. Referring to Figure 4 , the system adopts a three-layer architecture design, including a data acquisition layer, an algorithm processing layer, and an application service layer.

[0161] At the data acquisition layer, the system acquires various types of data required for scheduling optimization through multi-source perception. The GPS positioning subsystem uses high-precision RTK-GPS technology to achieve centimeter-level positioning and is compatible with multiple satellite navigation systems such as Beidou, GPS, and GLONASS, which are used to collect real-time agricultural machinery position information. The IoT sensor network is deployed on farmland and machinery with multiple types of sensors, including: mechanical sensors for monitoring engine status, fuel consumption, and working depth; image sensors for monitoring work quality. The communication network adopts a hierarchical structure, with LoRa and ZigBee used for short-range sensor data transmission, 4G / 5G mobile networks used for medium-range communication, and satellite communication used for long-range communication in remote mountainous areas. The system provides standardized data acquisition interfaces that support multiple data formats such as JSON, XML, and CSV to achieve compatibility and integration with third-party systems.

[0162] At the algorithm processing layer, the system completes data processing and optimization calculation. The data preprocessing module is used for data cleaning, format unification, and coordinate system conversion, and improves data quality through feature extraction and multi-source data fusion. The clustering analysis module implements a multi-constrained density peak clustering algorithm, supports dynamic parameter adjustment, clustering quality evaluation and visualization, and allows manual intervention for correction. The optimization solving module implements the core logic of the MEGWO algorithm, supports parallel computing to improve optimization efficiency, provides multiple termination condition settings and online parameter tuning functions. The decision support module is used for multi-scenario comparison, sensitivity and robustness analysis, risk assessment, and uncertainty handling, and provides a visualization interface to assist decision-making.

[0163] At the application service layer, the system provides user interaction interfaces and business logic functions. The web application system uses responsive design, supports PC and mobile access, and integrates GIS services (including Baidu Map and Gaode Map) to achieve map visualization, real-time monitoring, and automatic report generation. The mobile application supports Android and iOS platforms, with functions such as offline maps, caching, voice interaction, push notifications, and on-site photo uploading. API services provide RESTful interfaces, GraphQL queries, and Webhook callback mechanisms, and use OAuth2.0 for authentication and authorization. In terms of integrated interfaces, the system can interface with ERP systems, farm management systems, financial payment platforms, and insurance claims systems to achieve cross-system business collaboration.

[0164] In terms of system architecture features, the system has high availability, security, and performance optimization capabilities. High availability is achieved through microservice architecture, supporting independent service deployment, load balancing, failover, and data backup and disaster recovery. Security is guaranteed through a multi-layer protection mechanism, including data transmission and storage encryption, identity authentication and permission management, audit logs, and security monitoring. In terms of performance optimization, the system uses database indexing and query optimization, caching mechanisms, asynchronous processing, and CDN acceleration to improve response speed and concurrent processing capacity.

[0165] Embodiment 3:

[0166] Referring to Figure 5 and Figure 6 , this embodiment takes the actual application of a mountainous area agricultural machinery cooperative union in Guang'an City, Sichuan Province as an example to verify the feasibility of the method. The union is composed of five agricultural machinery cooperatives located in Jinshui County, Huayingshan City, and Qianfeng District, with complex terrain and an elevation of 600 to 1400 meters.

[0167] The region has more than 2800 hectares of farmland, scattered in 265 different plots. The farmland area ranges from 0.3 hectares to 12 hectares, with an average of about 10.5 hectares per plot. Topographic analysis shows that the average slope of the entire region is 13.2 degrees, with a maximum elevation difference of 780 meters, and 127 plots have a slope of more than 15 degrees.

[0168] The union has 78 types of agricultural machinery equipment, including 25 tractors of different power, 18 seeders, 20 harvesters, and 15 plant protection machines. In the spring of 2023, the union undertook a total of 1350 cubic meters of fertilization tasks, which needed to be completed within a 30-day time window from March 15 to April 15.

[0169] Data collection was carried out using the method, and all farmland plots were positioned using GPS measurement technology with an error of less than 0.8 meters. A digital elevation model was established using unmanned aerial vehicle aerial survey technology with a resolution of 2 meters x 2 meters. In the data preprocessing stage, 12 coordinate outliers were corrected and 8 missing area data were completed.

[0170] The multi-constrained density peak clustering algorithm divides the 265 plots of farmland into 5 clusters. Cluster A contains 54 plots of farmland with a total demand of 285 cubic meters and an average slope of 11 degrees. Cluster B covers 61 plots of farmland with a total demand of 312 cubic meters and an average slope of 8 degrees. Cluster C contains 49 plots of farmland with an average slope of 16 degrees and a total demand of 243 cubic meters. Clusters D and E contain 58 and 43 plots of farmland, respectively, with work demands of 289 cubic meters and 221 cubic meters, respectively.

[0171] The memory-enhanced grey wolf optimization algorithm found the first feasible solution at the 12th generation and reached the convergence state at the 28th generation. The entire optimization process took about 2 hours. The final generated scheduling scheme specified the job route, time arrangement and task allocation of each device.

[0172] Compared with the manual scheduling scheme, the total scheduling cost decreased from 873,000 yuan to 698,000 yuan, with a decrease of 20.1%. The job completion time was shortened from 26 days to 23 days. The mechanical utilization rate increased from 68.4% to 84.7%. Fuel consumption decreased by 1850 liters.

[0173] In the comparison test with other algorithms, the cost was reduced by 12.3% compared with the genetic algorithm, the time was shortened by 8.6% compared with the particle swarm algorithm, and the algorithm convergence speed was increased by 2.1 times.

[0174] During the implementation process, continuous rainfall from March 25 to 27 affected the original work plan of 35 farmlands. The system completed the scheme re-optimization the day after the rainfall began, postponed the work of the affected farmlands, and prioritized the arrangement of well-drained plots. According to the statistics, the actual loss was 3.2%.

[0175] After the end of the project, the survey showed that 91.5% of the farmers were satisfied with the scheduling scheme. The cooperative records showed that the scheduling work time was shortened from 2-3 days to the same day.

[0176] Through the actual application verification of the above various scenes, the effectiveness, practicality and advancement of the method and system of the application are fully proved, which provides important technical support for the intelligent development of modern agriculture.

[0177] Compared with the prior art, the application has the following beneficial effects:

[0178] Multi-objective collaborative optimization: Breakthrough the limitation of traditional single-objective optimization, consider cost minimization and time optimization at the same time, provide multiple balance schemes for users through Pareto frontier analysis, meet the needs of different application scenarios.

[0179] Intelligent clustering grouping: Propose multi-constraint density peak clustering algorithm, comprehensively consider multiple factors such as geographical distance, terrain condition and work capacity, realize intelligent grouping of farmland, and significantly reduce problem scale and solution complexity.

[0180] Algorithm performance enhancement: MEGWO algorithm effectively solves the premature convergence and local optimum problem of traditional GWO algorithm by integrating multiple improvement strategies, and significantly improves the solution quality and convergence speed.

[0181] Real-time dynamic adjustment: The system has real-time monitoring and dynamic adjustment capability, can timely respond to environmental changes and emergencies, and ensure the timeliness and adaptability of the scheduling scheme.

[0182] High system integration: Modular design and standardized interfaces are adopted to support seamless integration with existing farm management systems and IoT devices, reducing deployment costs and technical barriers.

[0183] In the description of the present specification, the description of the terms "one embodiment", "some embodiments", "an example", "a specific example", or "some examples" and the like means that the specific features, structures, materials or characteristics described in connection with the embodiment or example are included in at least one embodiment or example of the present application. In the present specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described can be combined in any one or more embodiments or examples in a suitable manner.

[0184] It is not difficult for those skilled in the art to understand that the present application includes any combination of the above description of the summary and the specific embodiment part and the parts shown in the drawings, limited by the length and in order to make the specification brief, each scheme formed by these combinations is not described one by one. Any modification, equivalent replacement, improvement, etc. made within the spirit and principles of the present application shall be included in the protection scope of the present application.

[0185] Although the embodiments of the present application have been shown and described above, it is understood that the above-described embodiments are exemplary and should not be construed as limiting the present application, and those skilled in the art can make changes, modifications, replacements and variations to the above-described embodiments without departing from the principles and spirit of the present application within the scope of the present application. The scope of the present application is defined by the appended claims and their equivalents.

Claims

1. A multi-objective agricultural machine scheduling method for cross-regional mountain farmland management, characterized in that, Comprise: Multi-modal data collection and preprocessing, including geographic coordinates, area information, job requirements, and time window constraints; Multi-constrained density peak clustering grouping, including applying a multi-constrained density peak clustering algorithm to divide the farmland plots and assign them as geographic clusters; Dual-objective optimization model construction, including establishing a dual-objective optimization model to minimize the total scheduling cost and total completion time, where the scheduling cost includes start-stop cost, travel distance cost, and penalty cost, and the completion time includes waiting time, transfer time, and job time; Memory-enhanced grey wolf optimization solution, including initializing the population of the memory-enhanced grey wolf optimization algorithm using a best point set-based initialization method to generate initial solutions; Non-dominated sorting and crowding distance selection, including using non-dominated sorting and crowding distance mechanisms to handle multi-objective functions; Grey wolf position updating and leader selection mechanism, including updating the position of the wolf pack under the guidance of alpha wolves, beta wolves, and delta wolves through an enhanced position updating formula, and calculating the fitness value of each individual; Implementing the memory mechanism by maintaining the exploration group and the memory group, where good solutions are stored and bad solutions are eliminated; Evolution operator integration, including introducing mutation, crossover, and selection operators based on the differential evolution strategy to improve search efficiency; Random local search and linear population reduction, including selecting some wolves in the memory group to perform random local search; Using a linear population size reduction strategy to dynamically reduce the population size based on fitness evaluation; Dynamic adjustment and real-time optimization of scheduling scheme, including generating the optimal scheduling path and resource allocation scheme, which clearly defines the allocation of each agricultural machine, the order of operations, and the operation time of each farmland plot.

2. The multi-objective agricultural machine scheduling method for cross-region mountain field management according to claim 1, wherein, The multi-constrained density peak clustering algorithm comprises: Calculating the density value of each farmland plot based on geographic coordinates and job constraints; Determining core points, boundary points, and noise points according to the field radius and minimum point number parameters; Identifying cluster centers in high-density areas; Expanding clusters around the cluster centers to group farmland plots that meet spatial and temporal constraints.

3. The multi-objective agricultural machine scheduling method for cross-region mountain field management according to claim 1, wherein, The best point set initialization comprises: Generating a standard point set in an s-dimensional Euclidean space by multiplying the basis function value of each dimension with the corresponding serial number to construct the point set coordinates; Calculating the deviation value to evaluate the distribution quality of the point set, which is determined by a specific mathematical relationship involving constants related to the geometric parameter g and the positive error parameter ε; Setting the geometric parameter g as a cosine function sequence based on the dimension serial number and a specific prime number p, where the dimension serial number ranges from 1 to s, and the prime number p satisfies the condition that it is not less than the dimension number s after subtracting 1.5, and the smallest prime number value that satisfies this condition is selected.

4. The multi-objective agricultural machine scheduling method for cross-region mountain field management according to claim 1, wherein, The memory mechanism comprises: Maintaining the exploration group and the memory group, and the size of the exploration group is equal to that of the memory group; After initialization, the best performing wolf is stored in the memory group; Updating the exploration group position in each iteration process and refreshing the memory group with improved solutions; Comparing the cost function values of the memory group and the exploration group leader wolf, and replacing the leader wolf when the memory group has a better solution.

5. The multi-objective agricultural machine scheduling method for cross-region mountain field management according to claim 1, wherein, The evolution operator comprises: The mutation operation is applied to the current optimal wolf alpha position as the benchmark, combined with a dynamic scaling factor and the difference between the optimal position and the random wolf position to generate a mutation vector; The scaling factor is dynamically adjusted, which is larger at the beginning of the algorithm to enhance global exploration ability, and linearly decreases to a smaller value with the iteration process; According to the preset crossover probability, the target vector and the mutation vector are subjected to a crossover operation, and the vector components are randomly selected to form a test individual; A greedy selection strategy is used to compare the fitness of the test vector and the target vector, and the individual with better fitness is preferentially retained as the next generation population member.

6. The multi-objective agricultural machine scheduling method for cross-region mountain field management according to claim 1, wherein, The random local search includes: Selecting the top half of individuals from the memory population according to the fitness to balance search efficiency and computational cost; Calculating the Euclidean distance between each selected individual and other individuals in the solution space to identify the nearest neighbor individual; According to the fitness of the neighbor individual, different position generation strategies are used, when the neighbor individual performs better, the new position is generated by interpolation method to approach it, when the neighbor individual performs worse, the new position is generated by extrapolation method to move away from it; The fitness of the generated new position is evaluated, and only when the fitness of the new position is better than the current position, the replacement is performed, otherwise the original position remains unchanged.

7. The multi-objective agricultural machine scheduling method for cross-region mountain field management according to claim 1, wherein, The linear population size reduction includes: Setting a larger initial population size at the beginning of the algorithm and a smaller minimum population size at the end of the algorithm; According to the ratio of the current function evaluation times to the maximum allowed evaluation times, the target population size of the next generation is linearly calculated; After each iteration is completed, the exploration population and the memory population are sorted in descending order according to the fitness, and the worst individual is identified; According to the target population size calculated, the individuals with the lowest fitness are eliminated from the end of the two types of populations, and the population size is gradually reduced to retain high-quality individuals.

Images