Photovoltaic cleaning robot scheduling method based on multi-agent cooperation

Abstract

The invention discloses a photovoltaic cleaning robot scheduling method based on multi-agent cooperation, and the method comprises the following steps: carrying out the region division, and generating a consistent reachability topological matrix; data are collected and processed to generate regional cleaning value parameters, and a regional attribute set is formed; an ant population and an ant lion population are initialized, and a multi-agent initial candidate scheduling scheme set is generated; executing random walk and restricted area transfer, and generating a multi-agent candidate scheduling scheme set; performing basic fitness calculation to generate a basic fitness value; detecting an area and channel path occupancy relation, and adding conflict punishment to form a collaborative fitness value set; generating a next-generation candidate scheduling scheme set meeting the reachability topological matrix and the cooperative behavior constraint according to the updating rule; and selecting a target scheduling scheme, and distributing the target scheduling scheme to the plurality of photovoltaic cleaning robots for cooperative execution. According to the method, multi-agent collaborative scheduling is realized by adopting improved ant lion optimization, and a reachable, efficient and conflict-free scheduling scheme is generated.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of intelligent scheduling, in particular to a photovoltaic cleaning robot scheduling method based on multi-agent cooperation. BACKGROUND

[0002] The cleaning operation of the existing photovoltaic power station is mostly dependent on a single or a small number of photovoltaic cleaning robots to perform independent operation, and the scheduling mode generally adopts a grid map or a simplified regional division method for path planning, lacks a reachability constraint matrix established for the complex structure of the photovoltaic array, and cannot accurately reflect the regional passage width, passage continuity and actual connectivity, thereby causing problems such as unexecutable path, passage congestion or cross-regional decision failure of the robot in a large-scale photovoltaic scene. At the same time, the traditional scheduling algorithm is mostly based on single-agent path generation logic, lacks a task decomposition and path combination mechanism that can support multi-robot cooperative execution, and it is difficult to ensure that multiple cleaning robots avoid mutual interference when running at the same time.

[0003] In the prior art, multi-robot scheduling usually adopts conventional modes such as ant colony algorithm and particle swarm algorithm, and its search process only relies on heuristic probability or local information, without considering the influence of photovoltaic array pollution difference on scheduling priority, nor establishing a cooperative judgment model for describing robot path conflict, regional overlap and passage occupation, so that multi-robot is prone to resource contention and scheduling conflict in cleaning task execution. At the same time, the existing swarm intelligence optimization algorithm does not form a cooperative fitness construction method for the photovoltaic cleaning scene, and cannot guarantee the overall running efficiency of the multi-robot through the path conflict suppression mechanism, thereby limiting the automation and efficiency level of the cleaning operation of the photovoltaic power station.

[0004] Therefore, how to provide a photovoltaic cleaning robot scheduling method based on multi-agent cooperation is a problem that those skilled in the art need to solve. SUMMARY

[0005] One object of the present application is to provide a photovoltaic cleaning robot scheduling method based on multi-agent cooperation, which uses an improved lion optimization to realize multi-agent cooperative scheduling and generate a reachable, efficient and conflict-free scheduling scheme.

[0006] According to the photovoltaic cleaning robot scheduling method based on multi-agent cooperation, the following steps are included:

[0007] Performing regional division on the photovoltaic array, generating a consistent reachability topology matrix based on the passage connectivity relationship between regions;

[0008] Collecting pollution data, performing quantitative processing, generating a regional cleaning value parameter, and constituting a regional attribute set together with the reachability topology matrix;

[0009] Initialize the ant colony and the lion colony, eliminate the areas that do not meet the channel connectivity condition, and generate the initial multi-agent candidate scheduling scheme set according to the initialization mode of the improved lion optimization algorithm;

[0010] Perform random walk, limit the area transfer, and generate the multi-agent candidate scheduling scheme set according to the random walk rule;

[0011] According to the area cleaning value parameter and the path length, the basic fitness of the multi-agent candidate scheduling scheme is calculated, and the basic fitness value is generated;

[0012] Based on the basic fitness value, the area occupation relationship and the channel path occupation relationship are detected, the conflict penalty is added, and the collaborative fitness value set is formed;

[0013] According to the update rule of the improved lion optimization algorithm, the next generation of candidate scheduling scheme set that meets the reachability topological matrix and collaborative behavior constraint is generated;

[0014] From the next generation of candidate scheduling scheme set, the multi-agent candidate scheduling scheme with the highest collaborative fitness value is selected as the target scheduling scheme, and is assigned to the multiple photovoltaic cleaning robots for collaborative execution.

[0015] Optionally, the generation of the consistent reachability topological matrix comprises:

[0016] According to the installation row direction and the installation column direction of the photovoltaic array, the photovoltaic array is regularly segmented to generate the region grid point set, and the row number and the column number of each region grid point are recorded;

[0017] According to the row number and the column number of the region grid point, the adjacent relationship between the region grid points is identified, and the adjacent region pairs are recorded as adjacent region pairs;

[0018] For each pair of adjacent region grid points, channel detection is performed, and when the detection result does not meet the passing condition, the adjacent region pair is recorded as an impassable relationship, and a passing relationship record is generated;

[0019] According to the row number and the column number of the region grid point, the matrix arrangement operation is performed on the passing relationship record, and the passable relationship and the impassable relationship between the region grid points are written in the corresponding position of the matrix in the fixed arrangement order, to generate the reachability topological matrix;

[0020] The passing consistency check is performed on the reachability topological matrix, the region grid point pair marked as passable in the matrix is checked in both directions, and the consistent reachability topological matrix is generated.

[0021] Optionally, the generation of the region attribute set comprises:

[0022] The pollution data collection is performed on all area grids of the photovoltaic array, and the collected data is recorded as a pollution original data set in sequence of the area grids;

[0023] According to the pollution original data set, scale processing is performed on the pollution coverage area, grade division processing is performed on the pollution thickness characteristics, and brightness grading processing is performed on the pollution color characteristics. The pollution data after the three types of processing are combined to form a pollution quantization input sequence in sequence of the area grids;

[0024] Based on the pollution quantization input sequence, the pollution degree of each area grid is quantized and calculated, and the quantization results of each area grid are recorded as a pollution quantization result set;

[0025] According to the pollution quantization result set, the area grids are cleaned and the value is converted. The pollution quantization results are divided into cleaning grades according to the preset pollution grade rule, a regional cleaning value parameter is generated, and the regional cleaning value parameters corresponding to all area grids are combined to form a regional cleaning value parameter set.

[0026] According to the row number and column number of the area grid, the regional cleaning value parameter set is arranged to form a regional cleaning value parameter sequence, and the regional cleaning value parameter sequence and the uniform accessibility topology matrix are combined to form a regional attribute set.

[0027] Optionally, the generation of the multi-agent initial candidate scheduling scheme set comprises:

[0028] According to the regional attribute set, the ant colony is initialized, the regional cleaning value parameter of each area grid in the regional attribute set is written into the initialization input sequence of the ant, and the initialization position candidate interval is assigned to the ant according to the order of the regional cleaning value parameter, and the ant initialization sequence set is formed.

[0029] According to the regional attribute set, the ant lion colony is initialized, and the unreachable area is removed according to the uniform accessibility topology matrix, and the ant lion initialization sequence set is generated.

[0030] The area grid positions in the ant initialization sequence set and the ant lion initialization sequence set are combined according to the row number and column number of the area grid to form a task path combination initialization input sequence corresponding to the multiple photovoltaic cleaning robots, and the path combination corresponding to the multiple robots is recorded as a multi-agent initialization path combination set.

[0031] According to the multi-agent initialization path combination set, the path positions without area connectivity condition are removed to form an accessible path combination set.

[0032] The path combination arrangement processing is performed based on the reachable path combination set according to the initialization mode of the improved ant lion optimization algorithm, and an initial arrangement sequence of the path combination satisfying the reachability constraint is obtained.

[0033] The initial arrangement sequence of the path combination is written into the multi-agent initial candidate scheduling scheme structure in a fixed order, and a multi-agent initial candidate scheduling scheme set is formed.

[0034] Optionally, the generation of the multi-agent candidate scheduling scheme set comprises:

[0035] The multi-agent path combination in each candidate scheduling scheme is executed from the multi-agent initial candidate scheduling scheme set as a starting point, and each region grid in the path combination is extracted in order as an initial sequence of the walk;

[0036] The migratable regions are screened according to the initial sequence of the walk and the uniform reachability topology matrix, and a set of migratable regions is generated;

[0037] On the basis of the set of migratable regions, a random walk of the multi-agent is executed, and the migratable positions are determined in order of the row number and column number of the region grid for each multi-agent path combination, and an initial walk path sequence is generated;

[0038] According to the random walk rule of the improved ant lion optimization algorithm, the initial walk path sequence is executed for walk adjustment processing, and a set of multi-agent walk path combinations is formed;

[0039] According to the set of multi-agent walk path combinations, the reachability of each region grid in the walk path is verified again by calling the uniform reachability topology matrix, and a set of reachable path sequences satisfying the reachability requirement is generated;

[0040] The set of reachable path sequences is written into the multi-agent candidate scheduling scheme structure in a fixed order, and a multi-agent candidate scheduling scheme set is formed.

[0041] Optionally, the generation of the basic fitness value comprises:

[0042] According to the set of multi-agent candidate scheduling schemes, the path sequence of the multi-photovoltaic cleaning robot in each candidate scheduling scheme is extracted, and the region grids in the path sequence are sequentially composed into a path analysis sequence;

[0043] According to the path analysis sequence, the path length of each photovoltaic cleaning robot is calculated, the path length value of each photovoltaic cleaning robot is generated, and all path length values are recorded as a path length set;

[0044] According to the path analysis sequence, the region cleaning value parameter of the region grid involved is extracted, and a path region cleaning value parameter sequence is formed.

[0045] Based on the path length set and the path region cleanliness value parameter sequence, a basic fitness calculation is performed on each candidate scheduling scheme to generate a basic fitness value corresponding to each candidate scheduling scheme.

[0046] Optionally, the formation of the cooperative fitness value set includes:

[0047] Based on the set of candidate scheduling schemes for multi-agents, the path sequences of multiple photovoltaic cleaning robots in each candidate scheduling scheme are extracted one by one, and the regional grid points in the path sequence are combined into a regional occupancy sequence according to the path order.

[0048] Based on the regional occupancy sequence, regional occupancy relationship detection is performed on each candidate scheduling scheme. Regional grid points in the regional occupancy sequence that are simultaneously occupied by multiple photovoltaic cleaning robots at the same time location are recorded as regional overlap records, forming a regional overlap detection list.

[0049] Based on the path sequence in the multi-agent candidate scheduling scheme, the channel path occupancy relationship is identified. Any two photovoltaic cleaning robots in the path sequence occupying the same channel or having path intersections within the same migration time period are recorded as channel conflict records, forming a channel conflict detection list.

[0050] For each candidate scheduling scheme, a conflict judgment is performed based on the regional overlap detection list and the channel conflict detection list. When there is a record in the regional overlap detection list or the channel conflict detection list, the candidate scheduling scheme is marked as a candidate scheduling scheme with a conflict, and the corresponding basic fitness value is read out as the input for conflict penalty calculation.

[0051] For candidate scheduling schemes with conflicts, perform conflict penalty processing, generate a fitness value after penalty processing, and record the fitness value after penalty processing as a cooperative fitness value;

[0052] For candidate scheduling schemes that do not have conflicts, their basic fitness values ​​are directly recorded as cooperative fitness values;

[0053] The collaborative fitness values ​​of all candidate scheduling schemes are arranged in a fixed order to form a set of collaborative fitness values.

[0054] Optionally, the generation of the next-generation candidate scheduling scheme set includes:

[0055] Based on the set of cooperative fitness values, the candidate scheduling schemes of the multi-agent system are sorted by fitness. The cooperative fitness values ​​are arranged in descending order, and the sorted cooperative fitness values ​​and the corresponding candidate scheduling schemes form a one-to-one fitness sorting sequence.

[0056] For candidate scheduling schemes of multi-agent agents whose collaborative fitness values ​​meet the preset retention conditions in the fitness ranking sequence, a scheme retention operation is performed, and they are recorded as a retained scheme sequence in the form of the original path combination.

[0057] For multi-agent candidate scheduling schemes whose collaborative fitness values ​​do not meet the retention conditions in the fitness ranking sequence, antlion encirclement input extraction is performed, and the path sequence corresponding to each candidate scheduling scheme is extracted as the encirclement input sequence.

[0058] The antlion location identifier is determined based on the set of cooperative fitness values. The path sequence corresponding to the multi-agent candidate scheduling scheme with the highest cooperative fitness value is taken as the antlion location sequence, and the antlion location sequence is taken as the path guidance benchmark.

[0059] For the encirclement input sequence without a retained scheme, path guidance update is performed according to the update rules of the improved antlion optimization algorithm to generate an updated path sequence, and the updated path sequence is recorded as a path update result set;

[0060] The reachability of all path sequences in the path update result set is verified. The grid points of the regions marked as impassable in the uniform reachability topology matrix are removed from the path sequences, generating a set of path sequences that satisfy the regional reachability constraints.

[0061] The reserved scheme sequence and the set of path sequences that satisfy the regional reachability constraints are written into the next-generation candidate scheduling scheme structure in a fixed order to form a set of next-generation candidate scheduling schemes that satisfy the consistent reachability topology matrix and cooperative behavior constraints.

[0062] Optionally, the photovoltaic cleaning robot collaboratively performs cleaning tasks including:

[0063] Based on the iterative control parameters, the improved antlion optimization algorithm is subjected to iterative termination condition judgment. When the current iteration number reaches the preset maximum iteration number, or the cooperative fitness value in the next generation candidate scheduling scheme set remains stable in continuous iteration, the iteration state is recorded as the termination state.

[0064] In the terminated state, all cooperative fitness values ​​are extracted from the next generation candidate scheduling scheme set, and the cooperative fitness values ​​are arranged in descending order to determine the multi-agent candidate scheduling scheme with the highest cooperative fitness value.

[0065] The candidate scheduling scheme with the highest cooperative fitness value is recorded as the target scheduling scheme, and the path combination in the target scheduling scheme is assigned to the corresponding photovoltaic cleaning robot according to the number order of the multiple photovoltaic cleaning robots.

[0066] The target path assigned by number is used as the final scheduling instruction, and multiple photovoltaic cleaning robots perform collaborative cleaning tasks according to the target scheduling scheme.

[0067] The beneficial effects of this invention are:

[0068] This invention constructs a consistent reachability topology matrix, enabling the accurate expression of regional connectivity relationships within a photovoltaic array in a structured and bidirectional manner. Multiple photovoltaic cleaning robots can directly define feasible migration paths based on this matrix during scheduling, fundamentally avoiding path inoperability issues caused by insufficient channel width, discontinuous passageways, or unreachable regions. The introduction of reachability topology constraints not only improves the reliability of path generation but also ensures the stability of scheduling and searching under large-scale array conditions, thereby significantly reducing the frequent path failures and redundant planning phenomena in traditional methods.

[0069] This invention quantifies and integrates pollution coverage area, pollution thickness, and color brightness to form a unified regional cleaning value parameter, which is then directly incorporated into the scheduling search. This allows multiple photovoltaic cleaning robots to prioritize cleaning based on differences in pollution levels during the task generation phase, thereby improving the overall rationality of cleaning task allocation. The regional attribute set, composed of the regional cleaning value parameter and the accessibility topology matrix, enables the scheduling process to proceed under both spatial constraints and task requirements. This allows robots to prioritize covering heavily polluted areas, reduce unnecessary path migration, and improve the cleaning efficiency of photovoltaic power plants.

[0070] This invention employs an improved antlion optimization algorithm, treating multi-agent path combinations as a unified optimization object to achieve joint optimization of path feasibility, pollution value, and robot collaborative behavior. Through mechanisms such as random walks, encirclement updates, and path guidance, multiple generations of candidate scheduling schemes continuously converge towards a better solution during the iteration process. By combining a set of collaborative fitness values, real-time conflict suppression is implemented for overlapping regions and intersecting paths, ensuring that multiple robots can avoid mutual interference during concurrent operation and guaranteeing the continuity and collaboration of the overall scheduling process. By introducing conflict penalty strategies and update rules, the scheduling results effectively suppress resource contention while ensuring coverage efficiency, thereby improving the overall execution efficiency of multi-robot collaborative cleaning tasks and the power generation revenue of photovoltaic power plants. Attached Figure Description

[0071] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings:

[0072] Figure 1 This is an overall flowchart of a photovoltaic cleaning robot scheduling method based on multi-agent collaboration proposed in this invention;

[0073] Figure 2 This is a schematic diagram illustrating the generation of a next-generation candidate scheduling scheme set based on the improved antlion optimization algorithm in the photovoltaic cleaning robot scheduling method based on multi-agent collaboration proposed in this invention. Detailed Implementation

[0074] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, illustrating only the basic structure of the invention, and therefore only show the components relevant to the invention.

[0075] refer to Figure 1 and Figure 2 A method for scheduling photovoltaic cleaning robots based on multi-agent collaboration includes the following steps:

[0076] The photovoltaic array is divided into regions, and a consistent reachability topology matrix is ​​generated based on the channel connectivity between regions to limit the feasible movement paths of the photovoltaic cleaning robot between regions.

[0077] Pollution data from each region of the photovoltaic array is collected, the pollution data is quantified, and regional clean value parameters are generated. These parameters, together with the reachability topology matrix, constitute a set of regional attributes for scheduling and searching.

[0078] The ant colony and antlion colony are initialized based on the set of regional attributes. The task paths corresponding to multiple photovoltaic cleaning robots are combined as multi-agent collaborative objects. The reachability topology matrix is ​​used to eliminate regions that do not have channel connectivity conditions. The initial candidate scheduling scheme set for multi-agents is generated according to the initialization method of the improved antlion optimization algorithm.

[0079] Starting from the initial set of candidate scheduling schemes, a random walk is performed. The regional transfer of multi-agent path combinations is restricted according to the reachability topology matrix, so that the ants can only change positions between regions with channel connectivity. A set of multi-agent candidate scheduling schemes that meet the reachability topology matrix requirements is generated according to the random walk rules of the improved antlion optimization algorithm.

[0080] Based on the regional cleaning value parameters and the path lengths of multiple photovoltaic cleaning robots in the multi-agent candidate scheduling scheme, the basic fitness of the multi-agent candidate scheduling scheme is calculated to generate a basic fitness value for conflict determination.

[0081] Based on the basic fitness value, the regional occupancy relationship and channel path occupancy relationship among multiple photovoltaic cleaning robots in the multi-agent candidate scheduling scheme are detected. For multi-agent candidate scheduling schemes with regional overlap or path intersection, conflict penalty is added to update the basic fitness value to the cooperative fitness value, and together with the basic fitness value of the candidate scheduling scheme without conflict, a cooperative fitness value set is formed.

[0082] Based on the cooperative fitness set value, antlion encirclement and path update are performed. Multi-agent candidate scheduling schemes that meet the cooperative fitness requirements are retained, and multi-agent candidate scheduling schemes that do not meet the cooperative fitness requirements are updated according to the antlion's position. The next generation candidate scheduling scheme set that meets the reachability topology matrix and cooperative behavior constraints is generated according to the update rules of the improved antlion optimization algorithm.

[0083] After the iteration termination condition is met, the multi-agent candidate scheduling scheme with the highest cooperative fitness value is selected from the next generation candidate scheduling scheme set as the target scheduling scheme, and the target scheduling scheme is assigned to multiple photovoltaic cleaning robots for collaborative execution.

[0084] In this embodiment, the generation of the uniform reachability topology matrix includes:

[0085] The photovoltaic array is regularly divided according to the installation row and column directions to generate a set of regional grid points with fixed row and column numbers, and the row and column numbers of each regional grid point are recorded.

[0086] Based on the row and column numbers of the regional grid points, the adjacency relationship between regional grid points is identified, and adjacent regional grid points are recorded as adjacent regional pairs, which are used to indicate that two regional grid points have directly adjacent positions in the array structure.

[0087] For each pair of adjacent grid points, a passage detection is performed. The detection includes whether there is a continuous physical passage space between the grid points and whether the passage space meets the passage width requirements of the photovoltaic cleaning robot. When the detection result meets the passage conditions, the adjacent area pair is recorded as a passable relationship. When the detection result does not meet the passage conditions, the adjacent area pair is recorded as an impassable relationship. A passage relationship record is generated to describe the passability between the grid points.

[0088] Based on the row and column numbers of the regional grid points, a matrix arrangement operation is performed on the accessibility records. Each regional grid point is used as the row index position and the column index position of the matrix. The accessible and inaccessible relationships between regional grid points are written into the corresponding positions of the matrix in a fixed arrangement order to generate an accessibility topology matrix. The accessibility topology matrix is ​​used to describe the accessibility relationship between any two regional grid points in the photovoltaic array. By detecting the channel width, channel continuity and adjacent positions of all regional grid points, the accessible and inaccessible relationships are expressed in matrix form to limit the feasible migration path of the photovoltaic cleaning robot between regions.

[0089] Perform a connectivity consistency check on the reachability topology matrix. Perform a bidirectional check on grid point pairs in the matrix that are marked as traversable in one direction. When two grid points meet the bidirectional connectivity condition, they are uniformly marked as bidirectionally traversable. When the bidirectional connectivity condition is not met, they remain as intraversable. Generate a consistent reachability topology matrix.

[0090] In this embodiment, the generation of the regional attribute set includes:

[0091] Pollution data was collected from all grid points of the photovoltaic array. The collected data included the pollution coverage area, pollution thickness characteristics, and pollution color characteristics. The collected data was recorded as the original pollution data set in the order of the grid points.

[0092] Based on the original pollution data set, the pollution coverage area is scaled, the pollution thickness characteristics are classified, and the pollution color characteristics are graded by brightness. The pollution data after the three types of processing are combined according to the regional grid order to form a pollution quantification input sequence.

[0093] Based on the pollution quantification input sequence, the pollution degree of each grid point in the region is quantified. The pollution degree is the sum of the scaled pollution coverage area and the graded pollution thickness, plus the brightness graded pollution color value. The quantification results of each grid point in the region are recorded as a pollution quantification result set.

[0094] Based on the pollution quantification result set, the regional grid points are converted into clean value. The pollution quantification results are divided into clean levels according to the preset pollution level rules, and regional clean value parameters are generated. The clean level value of the regional clean value parameter is multiplied by a fixed clean weight and then the regional number factor is added. The regional clean value parameters corresponding to all regional grid points are combined to form a regional clean value parameter set.

[0095] The set of regional cleaning value parameters is arranged according to the row and column numbers of the regional grid points to form a sequence of regional cleaning value parameters. The sequence of regional cleaning value parameters is then combined with the uniform accessibility topology matrix to form a set of regional attributes, so that the set of regional attributes contains both the regional cleaning value parameters and the accessibility constraint information between the regional grid points.

[0096] In this embodiment, the generation of the initial candidate scheduling scheme set for the multi-agent system includes:

[0097] The ant population is initialized based on the set of regional attributes. The regional cleanliness value parameter of each grid point in the set of regional attributes and the corresponding position of the uniform reachability topology matrix are written into the ant's initialization input sequence. Candidate intervals for initialization positions are assigned to the ants according to the order of the regional cleanliness value parameters, forming a set of ant initialization sequences.

[0098] The antlion population is initialized based on the set of regional attributes. The cleanliness value parameters of each grid point in the set of regional attributes are written into the antlion initialization input sequence in a fixed order. Unreachable regions are removed from the antlion initialization range based on the uniform reachability topology matrix, generating a set of antlion initialization sequences.

[0099] The region grid points in the ant initialization sequence set and the antlion initialization sequence set are combined according to the row number and column number of the region grid points to form the task path combination initialization input sequence corresponding to multiple photovoltaic cleaning robots, and the path combination corresponding to multiple robots is recorded as the multi-agent initialization path combination set.

[0100] Based on the multi-agent initial path combination set, path locations that do not meet the regional connectivity conditions are eliminated. The elimination rule is based on the reachability identifier in the consistent reachability topology matrix. When any regional location in the multi-agent initial path combination is marked as unreachable in the consistent reachability topology matrix, the path location is removed from the combination, forming a reachable path combination set.

[0101] Based on the set of reachable paths, the path combination arrangement process is performed according to the initialization method of the improved antlion optimization algorithm. The path combination arrangement process is to sort the multi-agent path combinations according to the regional cleaning value parameter, and then perform sequential cross arrangement in combination with the set of reachable paths to generate the arrangement result, so as to obtain the initial arrangement sequence of path combinations that satisfies the reachability constraint.

[0102] The initial permutation sequence of path combinations is written into the multi-agent initial candidate scheduling scheme structure in a fixed order to form a multi-agent initial candidate scheduling scheme set.

[0103] In this embodiment, the generation of the multi-agent candidate scheduling scheme set includes:

[0104] Starting from the initial set of candidate scheduling schemes for multi-agent agents, the walk input extraction is performed on the multi-agent path combination in each candidate scheduling scheme, and the grid points of each region in the path combination are extracted in sequence as the initial walk sequence.

[0105] Based on the initial walk sequence and the unified reachability topology matrix, the migrateable regions are filtered. When any region grid point in the initial walk sequence is marked as unreachable in the unified reachability topology matrix, the region grid point is removed from the migrateable region list, generating a set of migrateable regions.

[0106] Based on the set of transferable regions, a multi-agent random walk is performed. For each multi-agent path combination, the transferable position is determined sequentially according to the row and column numbers of the region grid points, generating an initial walk path sequence. The initial walk path sequence is advanced along the execution order of adjacent region grid points in the set of transferable regions.

[0107] The random walk process for generating the initial path sequence specifically includes: taking the starting region of each photovoltaic cleaning robot in the initial candidate scheduling scheme set of the multi-agent system as the starting point of the random walk; during the random walk, taking the current region as the reference, randomly selecting a region from the set of regions reachable by the current region as the next path location; the selection method for the next path location is as follows: in each migration step, randomly selecting one region from the set of traversable regions of the current region with equal probability as the next path region; after each region is selected, appending the selected region to the end of the path sequence of the corresponding photovoltaic cleaning robot, and using this region as the starting point of the next random walk; for regions in the random selection results that do not meet the region connectivity requirements, directly discarding the result and re-performing the random region selection operation to ensure that the path sequence always maintains regional continuity; repeating the above random selection and path appending process for each photovoltaic cleaning robot until the predetermined number of walk steps or path length limit is reached; finally, combining the path sequences formed by each photovoltaic cleaning robot through random walks in a fixed order to form the initial path sequence.

[0108] Based on the random walk rule of the improved antlion optimization algorithm, the initial walk path sequence is adjusted by sorting the regional grid points in the initial walk path sequence from largest to smallest according to the regional cleanliness value parameter, and then exchanging the order of adjacent regional grid points to generate the adjusted walk path sequence, thus forming a multi-agent walk path combination set.

[0109] Based on the set of multi-agent walk paths, the consistent reachability topology matrix is ​​called again to verify the reachability of the regional grid points in each walk path. When a regional grid point in the walk path sequence is marked as impassable at the corresponding position in the consistent reachability topology matrix, the regional grid point is removed from the path sequence to generate a set of reachable path sequences that meet the reachability requirements.

[0110] The set of reachable path sequences is written into the multi-agent candidate scheduling scheme structure in a fixed order to form a multi-agent candidate scheduling scheme set.

[0111] In this embodiment, the generation of the basic fitness value includes:

[0112] Based on the multi-agent candidate scheduling scheme set, the path sequences of multiple photovoltaic cleaning robots in each candidate scheduling scheme are extracted, and the regional grid points in the path sequence are arranged in order to form a path analysis sequence.

[0113] The path length of each photovoltaic cleaning robot is calculated based on the path analysis sequence. The regional migration action between adjacent grid points in the path analysis sequence is used as the length calculation unit to generate the path length value of each photovoltaic cleaning robot. The path length value is obtained by multiplying the number of adjacent grid points by a fixed path distance coefficient and then summing them up. All path length values ​​are recorded as a path length set.

[0114] Based on the path analysis sequence, regional cleaning value parameters are extracted from the grid points in the involved areas. The regional cleaning value parameters corresponding to each grid point in the path analysis sequence are read out in the regional attribute set to form a path regional cleaning value parameter sequence.

[0115] Based on the path length set and the path area cleanliness value parameter sequence, a basic fitness calculation is performed on each candidate scheduling scheme to generate a basic fitness value corresponding to each candidate scheduling scheme. The basic fitness value is calculated by adding all area cleanliness value parameters in the area cleanliness value parameter sequence and then subtracting the sum of all path length values ​​in the path length set.

[0116] In this embodiment, the formation of the cooperative fitness value set includes:

[0117] Based on the set of candidate scheduling schemes for multi-agents, the path sequences of multiple photovoltaic cleaning robots in each candidate scheduling scheme are extracted one by one, and the regional grid points in the path sequence are combined into a regional occupancy sequence according to the path order.

[0118] Based on the regional occupancy sequence, regional occupancy relationship detection is performed on each candidate scheduling scheme. Regional grid points in the regional occupancy sequence that are simultaneously occupied by multiple photovoltaic cleaning robots at the same time location are recorded as regional overlap records, forming a regional overlap detection list to determine whether there is regional contention behavior.

[0119] Based on the path sequence in the multi-agent candidate scheduling scheme, the channel path occupancy relationship is identified. Any two photovoltaic cleaning robots in the path sequence occupying the same channel or having path intersections within the same migration time period are recorded as channel conflict records, forming a channel conflict detection list, which is used to detect whether there is path intersection behavior.

[0120] For each candidate scheduling scheme, a conflict judgment is performed based on the regional overlap detection list and the channel conflict detection list. When there is a record in the regional overlap detection list or the channel conflict detection list, the candidate scheduling scheme is marked as a candidate scheduling scheme with a conflict, and the corresponding basic fitness value is read out as the input for conflict penalty calculation.

[0121] For candidate scheduling schemes with conflicts, conflict penalty processing is performed. The conflict penalty processing is the base fitness value minus the penalty coefficient obtained by adding the number of overlapping regions and the number of path intersections, and the penalty processing fitness value is generated. The penalty processing fitness value is recorded as the cooperative fitness value.

[0122] For candidate scheduling schemes that do not have conflicts, their basic fitness values ​​are directly recorded as cooperative fitness values, so that they can participate in the update in subsequent steps with an unpenalized fitness state.

[0123] The collaborative fitness values ​​of all candidate scheduling schemes are arranged in a fixed order to form a collaborative fitness value set. This set is used as the input for the antlion encirclement and path update steps, so that the next step can perform retention or update operations on the candidate scheduling schemes based on the collaborative fitness values.

[0124] In this embodiment, the generation of the next-generation candidate scheduling scheme set includes:

[0125] Based on the set of cooperative fitness values, the candidate scheduling schemes of the multi-agent system are sorted by fitness. The cooperative fitness values ​​are arranged in descending order, and the sorted cooperative fitness values ​​and the corresponding candidate scheduling schemes form a one-to-one fitness sorting sequence.

[0126] For candidate scheduling schemes of multi-agent agents whose collaborative fitness values ​​meet the preset retention conditions in the fitness ranking sequence, a scheme retention operation is performed, and they are recorded as a retained scheme sequence in the form of the original path combination.

[0127] For multi-agent candidate scheduling schemes whose collaborative fitness values ​​do not meet the retention conditions in the fitness ranking sequence, antlion encirclement input extraction is performed, and the path sequence corresponding to each candidate scheduling scheme is extracted as the encirclement input sequence for subsequent path guidance updates;

[0128] The antlion location identifier is determined based on the set of cooperative fitness values. The path sequence corresponding to the multi-agent candidate scheduling scheme with the highest cooperative fitness value is taken as the antlion location sequence. The antlion location sequence is used as the path guidance benchmark to guide the path update direction of the unreserved scheme.

[0129] For the encirclement input sequence without a retained scheme, path guidance update is performed according to the update rules of the improved antlion optimization algorithm. The path guidance update is to move each regional grid point in the encirclement input sequence closer to the regional grid point corresponding to the antlion position sequence to generate an updated path sequence, and record the updated path sequence as a path update result set.

[0130] Based on the consistent reachability topology matrix, all path sequences in the path update result set are checked for reachability. The grid points of the regions marked as impassable in the consistent reachability topology matrix are removed from the path sequences, generating a set of path sequences that satisfy the regional reachability constraints.

[0131] The reserved scheme sequence and the set of path sequences that satisfy the regional reachability constraints are written into the next-generation candidate scheduling scheme structure in a fixed order to form a set of next-generation candidate scheduling schemes that satisfy the consistent reachability topology matrix and cooperative behavior constraints.

[0132] In this embodiment, the photovoltaic cleaning robot collaboratively performs cleaning tasks including:

[0133] Based on the iterative control parameters, the improved antlion optimization algorithm is subjected to iterative termination condition judgment. When the current iteration number reaches the preset maximum iteration number, or the cooperative fitness value in the next generation candidate scheduling scheme set remains stable in continuous iteration, the iteration state is recorded as the termination state.

[0134] In the terminated state, all cooperative fitness values ​​are extracted from the next generation candidate scheduling scheme set, and the cooperative fitness values ​​are arranged in descending order to determine the multi-agent candidate scheduling scheme with the highest cooperative fitness value.

[0135] The candidate scheduling scheme with the highest collaborative fitness value is recorded as the target scheduling scheme. The path combination in the target scheduling scheme is assigned to the corresponding photovoltaic cleaning robot according to the number order of the multiple photovoltaic cleaning robots, so that each photovoltaic cleaning robot obtains its corresponding path execution content.

[0136] The target path assigned by number is used as the final scheduling instruction, and multiple photovoltaic cleaning robots perform collaborative cleaning tasks according to the target scheduling scheme.

[0137] Example 1:

[0138] To verify the feasibility of this invention in practice, it was applied to the daily operation and maintenance of a large photovoltaic power station in Northwest China. With the continuous expansion of installed capacity, multiple photovoltaic cleaning robots need to simultaneously perform cleaning operations among hundreds or even thousands of rows of photovoltaic arrays. This region is located in a dry climate with frequent sandstorms, resulting in rapid accumulation of contaminants on the module surfaces, significantly impacting power generation efficiency. In existing cleaning methods, multiple robots often rely on manually specified or fixed paths for cleaning. This lack of effective scheduling and coordination mechanisms addresses the narrow passageways within the arrays, irregular connectivity between areas, and uneven distribution of contaminants, easily leading to problems such as passageway congestion, route duplication, and uneven cleaning coverage. This not only reduces cleaning efficiency but also increases robot travel distance and power station cleaning costs. The photovoltaic cleaning robot scheduling method based on multi-agent collaboration proposed in this invention is applied in this scenario to solve the problem of automated coordination of photovoltaic cleaning tasks under conditions of complex spatial structures and uneven contaminant distribution.

[0139] In the test area of ​​this photovoltaic power station, the array is arranged in a north-south direction with inconsistent row and column spacing, resulting in some areas with restricted access. To achieve automated coordination of cleaning tasks, this invention first divides the array into regions and generates a consistent reachability topology matrix based on actual access conditions. This matrix accurately describes whether each pair of regions is passable. Using this matrix, this invention can precisely limit the robot's migration path between regions, preventing the robot from entering inaccessible areas or getting stuck due to insufficient passage width. Compared to manual path planning, this method makes executable paths clearer and more reliable, ensuring that the scheduling algorithm operates under real physical conditions.

[0140] Before entering the scheduling calculation, this invention comprehensively collects data on the array's contamination status. Staff acquire component images under different lighting conditions using drones and fixed cameras, and extract contamination coverage area, dust thickness characteristics, and brightness variations through image processing. The collected data is then quantified to generate regional cleaning value parameters, reflecting the cleaning priority of each area. In the afternoon when actual lighting conditions are strong, comparing the cleaning value parameters of different areas reveals significant differences in contamination levels. This invention can dynamically prioritize robot tasks based on these differences, allowing cleaning robots to prioritize heavily contaminated areas and improve overall cleaning efficiency.

[0141] In the specific scheduling implementation, the system initializes the ant and antlion populations and uses the starting positions of multiple photovoltaic cleaning robots as the multi-agent collaborative objects. Through the constraints of a consistent reachability topology matrix, all candidate paths lacking regional connectivity are eliminated during initialization, ensuring that each initial path is physically executable. During the random walk phase, the robot paths advance gradually between regions according to connectivity relationships, with each path extension occurring within the topological constraints. The generated set of initial candidate paths undergoes multiple walk adjustments, gradually achieving higher pollution coverage efficiency.

[0142] During the scheduling process, each candidate path calculates its basic fitness based on path length and contamination coverage, and further undergoes region overlap detection and channel conflict detection. In the test area, two robots clashed in narrow channels; this method quickly identified and suppressed such unexecutable paths through a conflict penalty mechanism, ensuring a smooth scheduling process. Finally, through ranking of cooperative fitness values, path updates, and reachability checks, the system progressively generates next-generation candidate scheduling schemes, continuously optimizing robot paths.

[0143] The testing ran from morning to evening, executing multiple rounds of scheduling under varying lighting, wind speeds, and pollution accumulation conditions. Results showed that the scheduling paths generated by this method exhibited higher regional coverage balance and significantly reduced path crossings between robots. In high-density contaminated areas, robots were able to complete cleaning tasks according to the priority order of regional cleaning value parameters, significantly reducing redundant cleaning. In complex arrays, the consistent reachability topology matrix resulted in more rational robot movement paths, eliminating invalid actions caused by entering inaccessible areas. Ultimately, the power plant achieved higher array cleaning integrity within a day, reduced robot movement and waiting time, and improved overall cleaning efficiency.

[0144] This embodiment illustrates that the present invention is applicable not only to regularly arranged photovoltaic power plant arrays, but also to scenarios with complex structures and uneven pollution distribution. By combining reachability topology constraints, regional cleaning value parameters, and multi-agent cooperative algorithms, the present invention can intelligently generate executable, efficient, and conflict-free cleaning paths in actual operating environments. This enables photovoltaic cleaning robots to maintain orderly collaboration even during high-density operations, reducing human intervention in the scheduling process and improving the overall operational quality and cleaning efficiency of power generation facilities.

[0145] Table 1. Performance Comparison of Various Scheduling Methods and the Method of the Present Invention

[0146] Index Manual scheduling mode Scheduling mode based on greedy rule Conventional scheduling algorithm mode Scheduling method of the present application Cleaning coverage (%) 78.3 82.1 87.6 96.4 Average path length (number of area migrations) 198 171 152 109 Multi-robot conflict frequency (times / hour) 11 9 7 1 Response time in high pollution area (minutes) 34 28 23 12 Overall task completion time (minutes) 221 206 184 133 System scheduling calculation time (seconds) Not applicable 22 38 14

[0147] As shown in Table 1, the method of this invention significantly outperforms the other three scheduling methods in all six key indicators. Regarding cleaning coverage, manual scheduling achieves only 78.3%, greedy scheduling 82.1%, and conventional scheduling 87.6%, while the method of this invention reaches 96.4%, exceeding the conventional method by 8.8 percentage points and the manual method by 18.1 percentage points, indicating that the method of this invention can more fully cover the contaminated area. In terms of path efficiency, the manual method requires 198 migrations, the greedy method reduces this to 171, the conventional method requires 152, and the method of this invention further reduces it to 109, a reduction of 43 migrations compared to the conventional method and 89 migrations compared to the manual method, resulting in a significantly shorter path and higher movement efficiency. Regarding multi-robot conflicts, the manual method experiences 11 conflicts per hour, the greedy method 9, the conventional method 7, while the method of this invention experiences only 1. Compared to conventional scheduling, the number of conflicts is reduced by 6, a decrease of nearly 86%, demonstrating a significant collaborative advantage.

[0148] In terms of response time in highly polluted areas, manual response takes 34 minutes, greedy response takes 28 minutes, and conventional response takes 23 minutes, while this invention only takes 12 minutes, which is 11 minutes faster than conventional response, nearly doubling the response efficiency. In terms of overall task completion time, manual response takes 221 minutes, greedy response takes 206 minutes, and conventional response takes 184 minutes, while this invention shortens it to 133 minutes, which is 51 minutes less than conventional response and 88 minutes less than manual response, significantly improving task execution efficiency.

[0149] Regarding the system scheduling calculation time, the greedy method takes 22 seconds, the conventional method takes 38 seconds, and the present invention takes only 14 seconds, which is 24 seconds faster than the conventional method, improving the calculation efficiency by more than 60%.

[0150] Overall, the method of this invention is the best in six quantitative indicators: coverage, path length, number of conflicts, response speed, task time, and computation time. The improvement in each indicator is directly supported by the figures in the table, which fully demonstrates its efficiency and superiority in large-scale photovoltaic clean scheduling.

[0151] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A method for scheduling photovoltaic cleaning robots based on multi-agent cooperation, characterized in that, Includes the following steps: The photovoltaic array is divided into regions, and a consistent reachability topology matrix is ​​generated based on the channel connectivity between regions. Pollution data is collected, quantified, and regional clean value parameters are generated, which together with the accessibility topology matrix constitute a set of regional attributes. Initialize the ant colony and antlion colony, remove areas that do not have channel connectivity, and generate a set of initial candidate scheduling schemes for multi-agent agents according to the initialization method of the improved antlion optimization algorithm. Perform random walks, restrict regional transfers, and generate a set of candidate scheduling schemes for multi-agents according to the random walk rules; Based on the regional cleaning value parameters and path length, the basic fitness of the multi-agent candidate scheduling schemes is calculated, and basic fitness values ​​are generated. Based on the basic fitness value, the region occupancy relationship and the channel path occupancy relationship are detected, and a conflict penalty is added to form a set of cooperative fitness values; Based on the update rules of the improved antlion optimization algorithm, a set of next-generation candidate scheduling schemes that satisfy the reachability topology matrix and cooperative behavior constraints is generated; The multi-agent candidate scheduling scheme with the highest cooperative fitness value is selected from the next-generation candidate scheduling scheme set as the target scheduling scheme and assigned to multiple photovoltaic cleaning robots for collaborative execution.

2. The photovoltaic cleaning robot scheduling method based on multi-agent collaboration according to claim 1, characterized in that, The generation of the uniform reachability topology matrix includes: The photovoltaic array is regularly divided according to the installation row and column directions to generate a set of regional grid points, and the row number and column number of each regional grid point are recorded. The adjacency relationship between regional grid points is identified based on the row and column numbers of the regional grid points, and adjacent regional grid points are recorded as adjacent regional pairs; Perform channel detection on each pair of adjacent grid points. If the detection result does not meet the passage conditions, record the adjacent pair as an impassable relationship and generate a passage relationship record. Based on the row and column numbers of the regional grid points, a matrix arrangement operation is performed on the accessibility records. The accessible and inaccessible relationships between regional grid points are written into the corresponding positions of the matrix in a fixed arrangement order to generate an accessibility topology matrix. Perform a passability consistency check on the reachability topology matrix, and perform a bidirectional check on grid points in the matrix that are unidirectionally marked as passable, to generate a consistent reachability topology matrix.

3. The photovoltaic cleaning robot scheduling method based on multi-agent collaboration according to claim 1, characterized in that, The generation of the region attribute set includes: Pollution data is collected from all grid points of the photovoltaic array, and the collected data is recorded as a raw pollution data set in the order of the grid points. Based on the original pollution data set, the pollution coverage area is scaled, the pollution thickness characteristics are classified, and the pollution color characteristics are graded by brightness. The pollution data after the three types of processing are combined according to the regional grid order to form a pollution quantification input sequence. Based on the pollution quantification input sequence, the pollution level of each regional grid point is quantified, and the quantification results of each regional grid point are recorded as a pollution quantification result set. Based on the pollution quantification results set, the regional grid points are converted into clean value, the pollution quantification results are divided into clean levels according to the preset pollution level rules, regional clean value parameters are generated, and the regional clean value parameters corresponding to all regional grid points are combined to form a regional clean value parameter set. The set of regional cleaning value parameters is arranged according to the row and column numbers of the regional grid points to form a regional cleaning value parameter sequence. The regional cleaning value parameter sequence is then combined with the uniform accessibility topology matrix to form a set of regional attributes.

4. The photovoltaic cleaning robot scheduling method based on multi-agent collaboration according to claim 1, characterized in that, The generation of the initial candidate scheduling scheme set for the multi-agent system includes: The ant population is initialized based on the set of regional attributes. The regional cleanliness value parameter of each grid point in the set of regional attributes and the corresponding position of the uniform reachability topology matrix are written into the ant's initialization input sequence. Candidate intervals for initialization positions are assigned to the ants according to the order of the regional cleanliness value parameters, forming a set of ant initialization sequences. The antlion population is initialized based on the set of regional attributes, and unreachable regions are removed based on the uniform reachability topology matrix to generate a set of antlion initialization sequences. The region grid points in the ant initialization sequence set and the antlion initialization sequence set are combined according to the row number and column number of the region grid points to form the task path combination initialization input sequence corresponding to multiple photovoltaic cleaning robots, and the path combination corresponding to multiple robots is recorded as the multi-agent initialization path combination set. Based on the multi-agent initial path combination set, path locations that do not meet the regional connectivity conditions are eliminated to form a reachable path combination set; Based on the set of reachable path combinations, the path combination permutation process is performed according to the initialization method of the improved antlion optimization algorithm to obtain the initial permutation sequence of path combinations that satisfy the reachability constraints; The initial permutation sequence of path combinations is written into the multi-agent initial candidate scheduling scheme structure in a fixed order to form a multi-agent initial candidate scheduling scheme set.

5. The photovoltaic cleaning robot scheduling method based on multi-agent collaboration according to claim 1, characterized in that, The generation of the multi-agent candidate scheduling scheme set includes: Starting from the initial set of candidate scheduling schemes for multi-agent agents, the walk input extraction is performed on the multi-agent path combination in each candidate scheduling scheme, and the grid points of each region in the path combination are extracted in sequence as the initial walk sequence. Based on the initial walk sequence and the uniform reachability topology matrix, the migrateable regions are filtered to generate a set of migrateable regions; Based on the set of transferable regions, perform multi-agent random walks. For each multi-agent path combination, determine the transferable positions in sequence according to the row and column numbers of the region grid points to generate an initial walk path sequence. Based on the random walk rules of the improved antlion optimization algorithm, the initial walk path sequence is adjusted to form a set of multi-agent walk path combinations. Based on the set of multi-agent walk paths, the consistent reachability topology matrix is ​​called again to verify the reachability of the regional grid points in each walk path, generating a set of reachable path sequences that meet the reachability requirements; The set of reachable path sequences is written into the multi-agent candidate scheduling scheme structure in a fixed order to form a multi-agent candidate scheduling scheme set.

6. The photovoltaic cleaning robot scheduling method based on multi-agent collaboration according to claim 1, characterized in that, The generation of the basic fitness value includes: Based on the multi-agent candidate scheduling scheme set, the path sequences of multiple photovoltaic cleaning robots in each candidate scheduling scheme are extracted, and the regional grid points in the path sequence are arranged in order to form a path analysis sequence. The path length of each photovoltaic cleaning robot is calculated based on the path analysis sequence, generating a path length value for each photovoltaic cleaning robot, and all path length values ​​are recorded as a path length set. Based on the path analysis sequence, regional cleaning value parameters are extracted from the grid points in the involved areas to form a path regional cleaning value parameter sequence; Based on the path length set and the path region cleanliness value parameter sequence, a basic fitness calculation is performed on each candidate scheduling scheme to generate a basic fitness value corresponding to each candidate scheduling scheme.

7. The photovoltaic cleaning robot scheduling method based on multi-agent collaboration according to claim 1, characterized in that, The formation of the set of cooperative fitness values ​​includes: Based on the set of candidate scheduling schemes for multi-agents, the path sequences of multiple photovoltaic cleaning robots in each candidate scheduling scheme are extracted one by one, and the regional grid points in the path sequence are combined into a regional occupancy sequence according to the path order. Based on the regional occupancy sequence, regional occupancy relationship detection is performed on each candidate scheduling scheme. Regional grid points in the regional occupancy sequence that are simultaneously occupied by multiple photovoltaic cleaning robots at the same time location are recorded as regional overlap records, forming a regional overlap detection list. Based on the path sequence in the multi-agent candidate scheduling scheme, the channel path occupancy relationship is identified. Any two photovoltaic cleaning robots in the path sequence occupying the same channel or having path intersections within the same migration time period are recorded as channel conflict records, forming a channel conflict detection list. For each candidate scheduling scheme, a conflict judgment is performed based on the regional overlap detection list and the channel conflict detection list. When there is a record in the regional overlap detection list or the channel conflict detection list, the candidate scheduling scheme is marked as a candidate scheduling scheme with a conflict, and the corresponding basic fitness value is read out as the input for conflict penalty calculation. For candidate scheduling schemes with conflicts, perform conflict penalty processing, generate a fitness value after penalty processing, and record the fitness value after penalty processing as a cooperative fitness value; For candidate scheduling schemes that do not have conflicts, their basic fitness values ​​are directly recorded as cooperative fitness values; The collaborative fitness values ​​of all candidate scheduling schemes are arranged in a fixed order to form a set of collaborative fitness values.

8. The photovoltaic cleaning robot scheduling method based on multi-agent collaboration according to claim 1, characterized in that, The generation of the next-generation candidate scheduling scheme set includes: Based on the set of cooperative fitness values, the candidate scheduling schemes of the multi-agent system are sorted by fitness. The cooperative fitness values ​​are arranged in descending order, and the sorted cooperative fitness values ​​and the corresponding candidate scheduling schemes form a one-to-one fitness sorting sequence. For multi-agent candidate scheduling schemes whose cooperative fitness values ​​meet the preset retention conditions, a scheme retention operation is performed, and they are recorded as a retained scheme sequence in the form of the original path combination. For multi-agent candidate scheduling schemes whose cooperative fitness values ​​do not meet the retention conditions, antlion encirclement input extraction is performed, and the path sequence corresponding to each candidate scheduling scheme is extracted as the encirclement input sequence. The antlion location identifier is determined based on the set of cooperative fitness values. The path sequence corresponding to the multi-agent candidate scheduling scheme with the highest cooperative fitness value is taken as the antlion location sequence, and the antlion location sequence is taken as the path guidance benchmark. For the encirclement input sequence without a retained scheme, path guidance update is performed according to the update rules of the improved antlion optimization algorithm to generate an updated path sequence, and the updated path sequence is recorded as a path update result set; The reachability of all path sequences in the path update result set is checked. The grid points of regions marked as impassable in the consistent reachability topology matrix are removed from the path sequences, generating a set of path sequences that satisfy the regional reachability constraints. The reserved scheme sequence and the set of path sequences that satisfy the regional reachability constraints are written into the next-generation candidate scheduling scheme structure in a fixed order to form a set of next-generation candidate scheduling schemes that satisfy the consistent reachability topology matrix and cooperative behavior constraints.

9. A method for scheduling photovoltaic cleaning robots based on multi-agent collaboration according to claim 1, characterized in that, The photovoltaic cleaning robots collaboratively perform cleaning tasks including: Based on the iterative control parameters, the improved antlion optimization algorithm is subjected to iterative termination condition judgment. When the current iteration number reaches the preset maximum iteration number, or the cooperative fitness value in the next generation candidate scheduling scheme set remains stable in continuous iteration, the iteration state is recorded as the termination state. In the terminated state, all cooperative fitness values ​​are extracted from the next generation candidate scheduling scheme set, and the cooperative fitness values ​​are arranged in descending order to determine the multi-agent candidate scheduling scheme with the highest cooperative fitness value. The candidate scheduling scheme with the highest cooperative fitness value is recorded as the target scheduling scheme, and the path combination in the target scheduling scheme is assigned to the corresponding photovoltaic cleaning robot according to the number order of the multiple photovoltaic cleaning robots. The target path assigned by number is used as the final scheduling instruction, and multiple photovoltaic cleaning robots perform collaborative cleaning tasks according to the target scheduling scheme.

Images