Decentralized dynamic grouping and recombining algorithm

Abstract

The invention discloses a decentralized dynamic grouping recombination algorithm, which comprises the following steps: dividing a cluster into a plurality of groups, each group comprising a plurality of nodes, for each group, calculating a multi-dimensional resource coordination index of a single node, and evaluating the balance of resource distribution in the group through a group coordination entropy; when the grouping capability is reduced to be below a preset threshold value, triggering a recombination mechanism; in the recombination process, the node selection priority is dynamically adjusted according to the gap type of the group; applying homomorphic mapping constraint to ensure that the capability of the selected node is not lower than the current average level of the target group; if a certain group is imbalanced after being drawn and called, a dynamic feedback mechanism is triggered, and nodes are drawn and called from a global idle node pool or a low-priority group for supplement; evaluating the matching degree of the group of nodes and the target group, and calculating an adaptation score; and when the groups exceeding the preset proportion are recombined, triggering the cluster to stop the grouping mechanism, and switching to a maintenance state.

Description

Technical Field

[0001] The present invention relates to a recombination algorithm for a group of fire - fighting UAV clusters after grouping based on dynamic multi - dimensional collaborative entropy. Background Art

[0002] In the prior art, there are the following problems in the grouping management of autonomous cluster systems:

[0003] Defect of static grouping: Traditional hash bucketing or random grouping strategies do not consider the dynamic capabilities of nodes (such as power, attack power), which easily leads to unbalanced grouping resources.

[0004] Low recombination efficiency: Existing resource allocation algorithms (such as the ant colony algorithm based on load balancing) do not incorporate physical topology constraints, which may cause communication delays due to excessive distances between nodes.

[0005] Insufficient stability: Frequent recombination easily triggers "cascading collapse", and there is a lack of a global suppression mechanism for the number of recombinations.

[0006] Security risks: There is a lack of authentication when nodes migrate, which is easily penetrated by malicious nodes.

[0007] The group cooperation of the wolf pack algorithm only considers a single energy parameter, and the "own action resources + carried resources" two - parameter system proposed by the present invention has not been invented. Summary of the Invention

[0008] Aiming at the insufficient technical support for recombination after autonomous cluster grouping, the present invention proposes a recombination algorithm for a group of fire - fighting UAV clusters after grouping based on dynamic multi - dimensional collaborative entropy. After autonomous cluster grouping, each group acts according to the group and outputs capabilities. When a certain group as a whole loses its output capability, through the algorithm of the present invention, nodes are adjusted from the existing groups to recombine the groups.

[0009] The technical solutions adopted by the present invention are as follows:

[0010] A decentralized dynamic grouping recombination algorithm, including the following steps:

[0011] (1) Divide the cluster into multiple groups, each group includes multiple nodes. For each group, calculate the multi - dimensional resource collaboration index (MRCI) of a single node, and evaluate the balance of resource distribution within the group through the group collaboration entropy (GCE);

[0012] (2) When the group capability drops below a preset threshold, trigger the recombination mechanism;

[0013] (3) During the recombination process:

[0014] Dynamically adjust the node selection priority according to the gap type of the group;

[0015] Apply homomorphic mapping constraints to ensure that the capabilities of the selected nodes are not lower than the current average level of the target group;

[0016] If a group becomes unbalanced after nodes are drawn out, trigger the dynamic feedback mechanism to draw regulatory nodes from the global idle node pool or low-priority groups for supplementation;

[0017] (4) Evaluate the matching degree between the nodes in this group and the target group, and calculate the adaptation score;

[0018] (5) When regrouping occurs in more than the preset proportion of groups, trigger the cluster stop grouping mechanism and switch to the maintenance state.

[0019] Furthermore, the Multidimensional Resource Collaboration Index (MRCI) is used to quantify the comprehensive capabilities of nodes, including their own action resources, resource output capabilities, task demand adaptability, communication latency, and physical distance.

[0020] Furthermore, the calculation formula for the Multidimensional Resource Collaboration Index is:

[0021]

[0022] Where: Ei is the value of the own action resources of node i, normalized to a percentage value;

[0023] Ai is the resource output resource weight of node i;

[0024] γ is the task demand coefficient;

[0025] τ i is the communication latency of node i;

[0026] D i→g is the physical distance from node i to target g, calculated through GPS or UWB positioning data.

[0027] Furthermore, the task demand coefficient γ is calculated in real time by each node, and the formula is:

[0028] γ = 1 + value of resources that the node can output / task intensity.

[0029] Furthermore, the calculation formula for the grouping collaboration entropy is:

[0030]

[0031] Where:

[0032] MRCI k is the comprehensive ability MRCI vector of the kth node in group j;

[0033] is the vector sum of the MRCIs of all nodes in group j;

[0034] ln is the natural logarithm and is used for entropy calculation.

[0035] Furthermore, the judgment for triggering the recombination mechanism in step 2) includes:

[0036] Calculate the decay rate Δ_g of the multi-dimensional resource collaboration index:

[0037] Δ_g = initial total MRCI / current total MRCI

[0038] Calculate the node loss penalty term:

[0039] Node loss penalty term = 1 + initial number of nodes / number of lost nodes

[0040] Triggering condition: If Δ_g < α, it is determined that recombination is required; α is a preset threshold.

[0041] Furthermore, in step 3), the method for dynamically adjusting the node selection priority is:

[0042] When there is a resource output gap in the group, preferentially select nodes with high resource output capabilities;

[0043] When there is an action gap in the group, preferentially select nodes with high remaining power and short distance.

[0044] Furthermore, in step 3), the conditions for the homomorphic mapping constraint are:

[0045]

[0046] Where: is the MRCI of node i in other groups;

[0047] is the average value of the current MRCI of the recombination target group g;

[0048] λ is the homomorphic coefficient.

[0049] Furthermore, in step 3), the triggering condition of the dynamic feedback mechanism and the calculation method of the number of nodes to be compensated are:

[0050] (1) When a certain group has been regulated for nodes and its group collaboration entropy (GCE) exceeds the preset safety threshold θ, trigger the dynamic feedback mechanism;

[0051] (2) The calculation formula for the number of nodes to be compensated is:

[0052]

[0053] Where: GCE j is the collaboration entropy of group j;

[0054] θ is the safety threshold of the collaboration entropy GCE;

[0055] is the current number of nodes in group j;

[0056] is the ceiling function.

[0057] Furthermore, in step 4), the calculation method of the adaptation score is as follows: convert the multi-dimensional resource collaboration index (MRCI) of the node into a scalar, and the calculation formula is:

[0058]

[0059] W g : priority weight;

[0060] D max : the maximum allocation distance allowed by the system;

[0061] D i→g : the physical distance from node i to target g;

[0062] Sort according to the score Si, and the higher the score, the more likely the node will be selected first.

[0063] The present invention has the following beneficial effects:

[0064] Define the multi-dimensional resource collaboration index (MRCI) to quantify the coupling relationship between the comprehensive ability of the node and the battlefield situation. Design the group collaboration entropy (GCE) to dynamically evaluate the balance of resource distribution within the group. Construct a dynamic feedback mechanism and a reorganization pause mechanism to prevent system instability caused by excessive reorganization. Propose homomorphic mapping constraints and topological coherence rules to ensure the physical reachability and task capabilities of the reorganized groups.

[0065] The present invention can be applied to the following scenarios:

[0066] Military field: Adaptive formation reorganization of unmanned aerial vehicle / machine dog clusters in battlefield environments.

[0067] Disaster rescue: Dynamic scheduling of rescue robot clusters in case of communication interruption or equipment damage.

[0068] Industrial inspection: Task reallocation of multi-robot systems in complex factory building environments. BRIEF DESCRIPTION OF THE DRAWINGS

[0069] Figure 1 is the reorganization decision-making process.

[0070] Figure 2 is the group collaboration entropy (GCE) and dynamic feedback mechanism diagram. DETAILED DESCRIPTION OF THE INVENTION

[0071] The present invention will be further described below with reference to the accompanying drawings.

[0072] Scenario description: The mobile nodes of the robotic dogs act collectively. Facing a unified goal, they need to assemble and act in groups. The action resources of the nodes themselves and the output resources are the basic conditions for the action. When a special situation occurs somewhere and the robotic dogs need to carry resources to act, due to the relatively large target size, a single robotic dog cannot generate an effective output value. Therefore, the robotic dog cluster needs to be grouped, and tasks are executed in groups. At this time, the resources of the mobile nodes need to be calculated in two parts. The first part is the action resources (electricity) of the nodes themselves. The second part is various types of output resources carried (ammunition, drinking water, food, etc.). Similarly, the grouping and formation also need to be calculated in this way. In this scenario, grouping realizes the scale of action ability. As the event itself develops, the mobile nodes become disabled or unable to act due to the exhaustion of resources (their own or output), resulting in a decline in the grouping ability. However, the task needs to be output in groups. At this time, it is necessary to transfer mobile nodes from other groups to join the disabled group to restore the ability of the disabled group and continue to act.

[0073] This is very similar to when wolves act collectively. They are divided into many groups, each with its own task. If an accident occurs during the journey and an attack causes serious losses to a group, the wolves in other groups will make up for the group and restore the group's ability again.

[0074] The following is a detailed description of the technical solution of the present invention based on the implementation of the robotic dogs.

[0075] (1) Divide the robotic dog cluster into multiple groups, each group includes multiple robotic dog nodes. For each group, calculate the multi-dimensional resource coordination index (MRCI) of a single robotic dog node, and evaluate the balance of resource distribution within the group through the group coordination entropy (GCE).

[0076] The multi-dimensional resource coordination index (MRCI) is used to quantify the comprehensive ability of the node, including its own action resources, resource output ability, task requirement adaptability, communication delay, and physical distance.

[0077] (2) When the grouping ability drops below the preset threshold, trigger the reorganization mechanism;

[0078] (3) During the reorganization process:

[0079] Dynamically adjust the node selection priority according to the gap type of the group;

[0080] Apply the homomorphism mapping constraint to ensure that the ability of the selected node is not lower than the current average level of the target group;

[0081] If a group becomes unbalanced after being transferred, trigger the dynamic feedback mechanism, and draw nodes from the global idle node pool or low-priority groups for supplementation;

[0082] (4) Evaluate the matching degree between the nodes in this group and the target group, and calculate the adaptation score;

[0083] When more than the preset proportion of groups are reorganized, the cluster stops the grouping mechanism and switches to the maintenance state.

[0084] 1. Multi-dimensional Resource Collaboration Index (MRCI)

[0085] Definition: Used to quantify the comprehensive ability of a single robotic dog node, consisting of four-dimensional real-time parameters.

[0086] Formula:

[0087]

[0088] Ei: The value of the node i's own action resources, normalized to a percentage value (0 - 100%).

[0089] Ai: The resource output resource weight of node i, with a preset static attribute (e.g., ordinary robotic dog = 1.0, robotic dog carrying outputtable resources = 2.0).

[0090] γ: Task requirement coefficient, calculated in real time by the robotic dog node, formula:

[0091] γ = 1 + value of outputtable resources of the robotic dog / task intensity

[0092] Task intensity: Preset task level (0 - 10), value of outputtable resources of the robotic dog: Preset value (0 - 10).

[0093] For example, the outputtable resources are at level 5. γ = 1 + 10 / 5 = 3 > 1.

[0094] If the task intensity is greater than the resource output ability of the robotic dog, then γ will be greater than 1, which usually means the task is relatively difficult for the robotic dog and its resources may not be sufficient to meet the task requirements. On the contrary, if the resource output ability of the robotic dog is greater than the task intensity, then γ will be less than 1, indicating that the task is relatively easy and the robotic dog's resources are sufficient to handle the task.

[0095] τ i : Communication delay of node i (unit: millisecond), measured by the round-trip time of heartbeats.

[0096] D i→g : Physical distance from node i to target g (unit: meter), calculated from GPS or UWB positioning data.

[0097] After independent calculation of each parameter, a four-dimensional vector is formed. Then, standardization processing is carried out to normalize each dimension to the [0, 1] interval (e.g., Ei′ = Ei / 100). It is used for node ability comparison and allocation decision-making.

[0098] 2. Group Collaboration Entropy (GCE)

[0099] The group leader node calculates the balance of resource distribution within the group. The lower the entropy value, the more concentrated the resources. Group members notify the group leader of their own MRCI i If the group leader becomes disabled, the group elects a new leader. Each node in the group uses heartbeats to mark its online status and transmits MRCI if it is online i The group leader node calculates the resources of this group:

[0100]

[0101] MRCI k : The comprehensive ability MRCI vector of the k-th node in group j.

[0102] The vector sum of MRCI of all nodes in group j (summed separately for each dimension).

[0103] ln: The natural logarithm, used for entropy calculation.

[0104] 3. When a node in a certain group becomes disabled (ability value lower than the threshold)

[0105] The leader node of group g regularly evaluates the gap between the current ability and the initial ability of this group.

[0106]

[0107] The total current MRCI of group g (summed separately for each dimension).

[0108] The initial total MRCI of group g.

[0109] The number of nodes lost in group g.

[0110] The initial number of nodes in group g.

[0111] Calculation process:

[0112] Calculate the MRCI decay rate, which is the ratio of the current total MRCI to the initial total MRCI (take the mean after independent calculation for each dimension).

[0113] Add the node loss penalty term:

[0114] Node loss penalty term = 1 + initial number of nodes / number of lost nodes

[0115] Trigger condition: If Δ_g < α (for example, α = 0.6), it is determined that reorganization is required, and α is a preset threshold. The leader node of this group calculates the resources required for reorganization.

[0116] 4. Priority Weight Formula

[0117] The group leader calculates the missing resources of the group and dynamically adjusts the node selection priority according to the gap type.

[0118]

[0119] Robot dog action gap: the gap caused by insufficient power in the reorganization demand group (for example, remaining power < preset value).

[0120] Resource output gap: The gap caused by insufficient output resources of the reorganization demand group (for example, the remaining ammunition amount < the preset value).

[0121] The reciprocal of the distance from node i to target group g (i.e., the closer the distance, the larger the value).

[0122] Calculation process:

[0123] Select the weight formula according to the gap type. Resource output gaps focus on high-output resource robots, and action gaps focus on high-power and short-range robots. Output the priority score of each candidate node for sorting and selection. The leader node of this group sends a reorganization request and the required resource value to other group leader nodes.

[0124] 5. Homomorphic Constraint

[0125] After receiving the request, the leaders of other groups analyze the nodes in their group. value to ensure that the capacity of the selected nodes is not lower than the current average level of the target group.

[0126]

[0127] MRCI of node i in other groups.

[0128] Reorganize the mean of the current MRCI of the target group g (total MRCI / number of surviving nodes).

[0129] λ: Homomorphic coefficient (e.g. λ=1.2), to prevent inefficient nodes from being called in.

[0130] 6. Dynamic Compensation Mechanism

[0131] If a group is unbalanced after being adjusted, reverse compensation will be triggered. Number of nodes to be compensated

[0132]

[0133] GCE j : Co - entropy of group j.

[0134] θ: GCE security threshold (e.g., θ = 0.5).

[0135] Current number of nodes in group j.

[0136] Ceiling function.

[0137] Calculation process:

[0138] If GCE after transfer j > θ, calculate the excess ratio: (GCEj - θ) / θ. Determine the number of nodes to be compensated according to the ratio. Select adjustment nodes from the global idle node pool or low - priority groups as the supplementary source.

[0139] 7. Adaptation Score

[0140] Element definition:

[0141] W g : Priority weight (see details in 4).

[0142] D max : Maximum allowable deployment distance of the system (e.g., 200 meters).

[0143] D i→g : Physical distance from node i to target g (unit: meter), calculated through GPS or UWB positioning data.

[0144] Calculation process:

[0145] Convert the MRCI vector to a scalar:

[0146]

[0147] E i : Self - action resource value of node i.

[0148] A i : Resource output resource weight of node i.

[0149] τ i : Communication delay of node i.

[0150] D i→g : Physical distance from node i to target g.

[0151] Calculate the score S i ,

[0152]

[0153] MRCI i : Multi-dimensional resource synergy index of node i.

[0154] Distance attenuation item. The closer the distance, the larger the value.

[0155] The nodes are sorted according to the score Si. The higher the score, the higher the priority of being selected.

[0156] 8. Cluster stops grouping

[0157] When more than 2 / 3 of the groups are reorganized, the entire cluster stops grouping and is judged to be collectively disabled. Switch to the maintenance state and stop executing tasks to avoid losing more robot dog nodes. Each group leader node broadcasts the cluster stop command (for example, Type = 0x05PDU). All robot dog nodes stop task output, switch to the maintenance state (only heartbeat and basic communication) and return.

[0158] The above description is only a preferred embodiment of the present invention. It should be pointed out that a person skilled in the art can make several improvements without departing from the principle of the present invention, and these improvements should also be regarded as within the protection scope of the present invention.

Claims

1. A decentralized dynamic grouping and reorganization algorithm, characterized in that: The following steps are involved: (1) The cluster is divided into multiple groups, each of which includes multiple nodes. For each group, the multidimensional resource coordination index (MRCI) of a single node is calculated, and the balance of resource distribution within the group is evaluated by the group coordination entropy (GCE); (2) When the grouping capacity drops below a preset threshold, the reorganization mechanism is triggered; (3) During the reorganization process: Dynamically adjust node selection priority based on grouped gap types; Apply homomorphic mapping constraints to ensure that the capabilities of the selected nodes are not lower than the current average level of the target group; If a group becomes unbalanced after being transferred, the dynamic feedback mechanism is triggered, and the adjustment nodes are transferred from the global idle node pool or the low-priority group to supplement it; (4) Evaluate the matching degree between the nodes in this group and the target group and calculate the adaptation score; (5) When more than a preset ratio of groups are reorganized, the cluster stops grouping and switches to a maintenance state.

2. The decentralized dynamic grouping and reorganization algorithm according to claim 1, characterized in that: The Multidimensional Resource Coordination Index (MRCI) is used to quantify the comprehensive capabilities of a node, including its own action resources, resource output capability, adaptability to task requirements, communication delay, and physical distance.

3. The decentralized dynamic grouping and reorganization algorithm according to claim 1, wherein: The calculation formula of the multi-dimensional resource synergy index is: Where: Ei is the action resource value of node i, normalized to a percentage value; Ai is the resource output resource weight of node i; γ is the task demand coefficient; τ i is the communication delay of node i; D i→g It is the physical distance from node i to target g, calculated from GPS or UWB positioning data.

4. The decentralized dynamic grouping and reorganization algorithm according to claim 2, wherein: The task demand coefficient γ is calculated by each node in real time, formula: γ = 1 + node output resource value / task intensity.

5. The decentralized dynamic grouping and reorganization algorithm according to claim 1, characterized in that: The calculation formula of the group cooperative entropy is: in: MRCI k is the MRCI vector of the comprehensive ability of the k-th node in group j; The vector sum of MRCI for all nodes of group j; ln is the natural logarithm, which is used for entropy calculation.

6. The decentralized dynamic grouping and reorganization algorithm according to claim 1, characterized in that: The judgment of triggering the recombination mechanism in step 2) includes: Calculate the decay rate Δ_g of the multi-dimensional resource synergy index: Δ_g = initial total MRCI / current total MRCI Calculate the node loss penalty: Node loss penalty = 1 + initial number of nodes / number of lost nodes Trigger condition: if Δ_g<α, it is determined that reorganization is required; α is the preset threshold.

7. The decentralized dynamic grouping and reorganization algorithm according to claim 1, characterized in that: In step 3), the method for dynamically adjusting the node selection priority is: When there is a resource output gap in the group, nodes with high resource output capacity are given priority; When there is an action gap in the group, nodes with high remaining power and close distance are given priority.

8. The decentralized dynamic grouping and reorganization algorithm according to claim 1, characterized in that: In step 3), the conditions for the homomorphic mapping constraints are: Wherein: is the MRCI of node i in other groups; is the mean of the current MRCI for the recombination target group g; λ is the homomorphism coefficient.

9. The decentralized dynamic grouping and reorganization algorithm according to claim 1, characterized in that: In step 3), the triggering conditions of the dynamic feedback mechanism and the calculation method of the number of nodes to be compensated are: (1) When a group is selected for adjustment and its group cooperative entropy (GCE) exceeds the preset safety threshold θ, the dynamic feedback mechanism is triggered; (2) The calculation formula for the number of nodes to be compensated is: where: GCE j is the collaborative entropy of group j; θ is the collaborative entropy GCE safety threshold; is the current number of nodes for group j; is the ceiling function.

10. The decentralized dynamic grouping and reorganization algorithm according to claim 1, characterized in that: In step 4), the calculation method of the adaptation score is: convert the multidimensional resource coordination index (MRCI) of the node into a scalar, and the calculation formula is: W g : Priority weight; D max : The maximum allocation distance allowed by the system; D i→g : The physical distance from node i to the target g; The nodes are sorted according to the score Si. The higher the score, the higher the priority of being selected.

Images