A multi-intelligent node driven cluster decision method, system, device and medium

Abstract

This invention discloses a multi-intelligent node-driven cluster decision-making method, system, device, and medium, relating to the field of computer platform load balancing technology. The method includes broadcasting node operating parameters in the initial state of nodes and electing a master node; generating a dataset of node task awareness results based on the master node; calculating task complexity coefficients and node task experience weighting factors based on the dataset of node task awareness results; executing the convergence of the master node's state and generating a multi-task consensus decision vector with a weighted voting mechanism; and initiating a redundant master election process through a preset heartbeat judgment mechanism after detecting master node failure, and completing consensus updates under the new master node. The method described in this invention achieves dynamic adaptive scheduling of task allocation by constructing a weighted task scoring mechanism based on task complexity coefficients and experience factors, improving the rationality of task scheduling and the flexibility of system response.

Description

Technical Field

[0001] This invention relates to the field of intelligent cluster control and multi-agent autonomous collaboration technology, specifically to a method, system, device, and medium for multi-intelligent node-driven cluster decision-making. Background Technology

[0002] With the rapid development of intelligent technologies, collaborative control based on multi-agent systems has been widely applied in scenarios such as drone swarms, unmanned vehicle formations, unmanned surface vessel formations, and collaborative operations of intelligent robots. In these systems, how to achieve task coordination, state awareness, decision consensus, and fault tolerance among multiple intelligent nodes is the core issue for ensuring stable system operation and efficient task execution.

[0003] Most existing cluster control schemes rely on centralized scheduling strategies, where a central master node uniformly allocates tasks and issues commands. While this approach offers a clear structure and high control efficiency, it faces significant bottlenecks in practical deployments: if the master node fails or communication is interrupted, the entire system will be forced to halt, preventing tasks from completing. Furthermore, centralized architectures struggle to adapt to dynamic environmental changes and lack flexibility and self-adaptability.

[0004] To address these issues, some studies have introduced distributed control models, achieving a certain degree of collaboration through autonomous perception and limited communication between intelligent nodes. However, these solutions typically lack a robust task consensus mechanism and fail to establish reasonable node scoring strategies and voting mechanisms, making them prone to scheduling chaos or even system-wide inconsistencies when task conflicts or node failures occur. Summary of the Invention

[0005] In view of the above-mentioned problems, the present invention is proposed.

[0006] Therefore, the technical problem solved by this invention is that existing multi-intelligent node cluster control methods rely on a central node, have rigid task scheduling, poor fault tolerance, and low task consensus efficiency. It also addresses how to achieve decentralized and rapid master node re-election after the master node fails, how to build a dynamic weighted consensus mechanism based on task complexity and node experience factors, and how to achieve network-wide consistency and recoverability of task allocation.

[0007] To address the aforementioned technical problems, this invention provides the following technical solution: a multi-intelligent node-driven cluster decision-making method, comprising broadcasting node operating parameters in the initial state of nodes and electing a master node, and generating a dataset of node task perception results based on the master node.

[0008] The dataset based on the node task perception results is used to calculate the task complexity coefficient and the node task experience weight factor. The state of the task master node converges and a multi-task consensus decision vector with a weighted voting mechanism is generated.

[0009] After detecting the failure of the master node, a redundant master election process is initiated through a preset heartbeat judgment mechanism, and consensus update is completed under the new master node.

[0010] The weighted scoring mechanism that integrates task complexity and node experience factors includes: nodes dynamically calculate task complexity coefficients based on task priority, execution time and remaining energy, and generate experience weights by combining historical completion rates. After being uploaded to the master node, the weights are scored, sorted and scheduled.

[0011] The master node includes a consensus vector constructed using a voting weight normalization strategy, an objection feedback and correction mechanism, and dynamic reconstruction and version management of the task consensus structure.

[0012] As a preferred embodiment of the multi-intelligent node-driven cluster decision-making method of the present invention, the master node election includes the node entering a broadcast state after startup and setting a random delay to send an initialization broadcast packet.

[0013] The initial broadcast packet includes the node number, battery percentage, communication quality (Qi), and current task buffer information.

[0014] Once a node receives at least a set number of broadcast packets, it enters the RAFT lightweight master election process. While retaining the master election phase, the log replication process is omitted, and nodes with task complexity scores below a preset threshold and guaranteed to have communication quality indicators above the average value are given priority as candidate masters.

[0015] As a preferred embodiment of the multi-intelligent node-driven cluster decision-making method described in this invention, the dataset for generating node task perception results includes a dataset formed by the master node converging task priority, execution time, remaining energy, and historical completion rate data through a method whereby child nodes broadcast local task status perception information packets to the master node, thereby forming a dataset based on node task perception results.

[0016] As a preferred embodiment of the multi-intelligent node-driven cluster decision-making method of the present invention, the computational task complexity coefficient and node task experience weight factor include: each sub-node generates a task complexity coefficient based on key real-time parameters of the dataset, and the weight coefficient is dynamically adjusted according to the environment.

[0017] The complexity coefficient is calculated by combining the weighted sum of the target priority and the estimated execution time with the inverse ratio of the remaining power.

[0018] The completion status of each node in the recent task execution process is statistically analyzed, and the task completion rate of the node is calculated. The completion rate is used as the experience weight factor of the node, and the stability and reliability of the node's task execution are quantified through the experience weight.

[0019] The task complexity coefficient and empirical weight factor are uploaded to the master node through a state broadcasting mechanism to achieve state convergence of the task master node.

[0020] As a preferred embodiment of the multi-intelligent node-driven cluster decision-making method described in this invention, the state convergence of the execution task master node includes, after the master node receives the task complexity coefficients and empirical factors reported by all child nodes, organizing the state information of all nodes into a unified data structure and caching the fields in matrix form.

[0021] The fields cached include the node's unique ID, node task complexity coefficient, node experience weight factor, node communication quality score, and node task load.

[0022] The master node ranks and scores all candidate nodes corresponding to each scheduled task according to a unified scheduling and scoring logic. The scoring criteria are constructed using the inverse relationship between the node's experience factor and the task complexity coefficient.

[0023] The master node selects the node with the highest score to execute the task first, and generates a task number, assigns a node number, and assigns a scheduling score as the task scheduling scheme.

[0024] The task scheduling scheme is integrated into a set of weighted task consensus vectors, which are marked with version numbers by the master node and broadcast to all nodes to enter the consensus confirmation phase.

[0025] During the consensus confirmation process, if node feedback indicates disagreements regarding some task allocations, the master node will dynamically reconstruct the consensus vector based on the feedback suggestions and rebroadcast it until a stable and consistent task allocation structure is formed.

[0026] As a preferred embodiment of the multi-intelligent node-driven cluster decision-making method of the present invention, the generation of the multi-task consensus decision vector with weighted voting mechanism includes: after the master node completes the convergence of the task status, it constructs a multi-task weighted scheduling and scoring model based on the task complexity coefficient and task completion history factor uploaded by each child node.

[0027] The master node calculates the voting weight score of the node in the current scheduling round based on the task complexity coefficient and task completion history rate of the child node.

[0028] The voting weight score is determined by the inverse relationship between the node's experience factor and the task complexity coefficient. The higher the node's score, the greater its weight in the voting mechanism.

[0029] For each task to be scheduled, the master node selects the nodes with higher voting weight scores from all candidate nodes of executable tasks as the candidate set.

[0030] The master node normalizes the voting weights of each node in the candidate set and constructs a weight vector structure for task allocation.

[0031] The vector structure consists of triplets, namely the task number, the candidate node number, and the weighted score value of the corresponding node.

[0032] As a preferred embodiment of the multi-intelligent node-driven cluster decision-making method described in this invention, the generation of a multi-task consensus decision vector with a weighted voting mechanism further includes the master node prioritizing the determination of the task scheduling node based on the weighted score value, and combining the task number with the optimal node number to generate a preliminary task consensus decision vector.

[0033] The task consensus decision vector is broadcast to all child nodes in the form of a structured data packet, and a consistency confirmation window is opened. Each child node returns confirmation and objection feedback flags based on the received allocation results and its own task awareness.

[0034] The master node adjusts the weights of conflicting task assignments based on the feedback data returned by the child nodes. It uses a feedback frequency-weighted reconstruction method to dynamically correct the task weight values ​​of the corresponding nodes and forms a revised consensus vector marked with a version number.

[0035] When the consensus feedback rate reaches a preset threshold, the master node solidifies the current consensus decision vector version as the task allocation and execution benchmark within the scheduling cycle, and distributes it to all nodes for synchronous execution.

[0036] As a preferred embodiment of the multi-intelligent node driven cluster decision-making method of the present invention, the step of initiating the redundant master election process through a preset heartbeat judgment mechanism includes: all non-master nodes setting a fixed master control heartbeat listening period, with a default value of 10 seconds, and attempting to receive the heartbeat signal broadcast by the master control node in each listening period. The heartbeat signal is a predefined short status synchronization packet, which includes the master control node number, task consensus version number, broadcast timestamp, and master control running status summary.

[0037] Each child node is equipped with a packet loss counter. If it fails to receive a heartbeat packet from the master node for three consecutive heartbeat cycles, it will automatically mark the master node as suspected of failure and enter the local judgment waiting stage.

[0038] During the local determination phase, the node will attempt to send a master control activity confirmation packet to the master control node via unicast. If there is still no response to the confirmation packet within three consecutive heartbeat cycles, the master control status will be officially set to unavailable, and a master control failure notification message will be broadcast in the network.

[0039] Upon receiving the master controller failure notification, all nodes immediately cease their original consensus scheduling operations and uniformly enter the candidate master controller information broadcasting phase, where each node broadcasts its own master controller candidate status packet.

[0040] The broadcast's own master control candidate status packet includes the node number, the current task cache quantity and total complexity index, the remaining power percentage, the current communication quality index, and the node's historical task completion rate.

[0041] After broadcasting, all nodes wait to receive candidate information from other nodes. After the statistics are completed, each node will independently calculate the score based on a unified comprehensive scoring function. The node with the highest score waits for the master control confirmation packet to be broadcast after the broadcast round ends and requests master control confirmation feedback from the entire network.

[0042] The broadcast master confirmation packet includes its own score, version number, and candidate proof packet summary.

[0043] If there are no objections after receiving the response, a confirmation response is returned. When the node responses are consistent, the node officially switches to the new master state. The new master node inherits the consensus task cache structure of the previous master and immediately starts the state recovery and task reconstruction process.

[0044] As a preferred embodiment of the multi-intelligent node driven cluster decision-making method described in this invention, the step of completing the consensus update under the new master node includes loading the latest task consensus vector and version number from the original master node broadcast cache after the new master node confirms the succession, and performing consistency verification on the status of all incomplete tasks.

[0045] When the node feedback consistency ratio exceeds the set threshold, the new master node confirms the validity of this round of revision consensus, officially enters the task scheduling and synchronous execution process, and increments the consensus version number by 1 to record it as the current task version.

[0046] If no consensus is reached, the master node will perform a single secondary revision based on the conflicting task number, redistribute tasks within a limited scope, and conduct up to N rounds of feedback iteration. If no effective consensus is still formed, the master node will use a forced instruction method to complete the task assignment and mark the downgraded consensus version to enter the downgraded operation state.

[0047] Another objective of this invention is to provide a multi-intelligent node-driven cluster decision-making system that can achieve multi-node scoring and dynamic negotiation before task allocation by constructing a weighted task consensus mechanism that integrates task complexity assessment and node experience weights. This solves the problems of single point of failure of the master node, rigid task allocation, and lack of adaptability of scheduling strategies that are common in current cluster scheduling systems.

[0048] As a preferred embodiment of the multi-intelligent node driven cluster decision-making system of the present invention, it includes a data acquisition and generation module, a decision calculation module, and a detection and update module.

[0049] The acquisition and generation module is used to broadcast the node's running parameters in the node's initialization state and to elect a master node, and to generate a dataset of node task perception results based on the master node.

[0050] The decision calculation module is used to calculate the task complexity coefficient and the node task experience weight factor based on the dataset of node task perception results, execute the state convergence of the task master node, and generate a multi-task consensus decision vector with a weighted voting mechanism.

[0051] The detection and update module is used to initiate a redundant master election process and complete the consensus update under the new master after the master node fails by using a preset heartbeat judgment mechanism.

[0052] A computer device includes a memory and a processor, the memory storing a computer program, and the processor executing the computer program to implement a multi-intelligent node-driven cluster decision-making method.

[0053] A computer-readable storage medium having a computer program stored thereon, the computer program implementing steps of a multi-intelligent node-driven cluster decision-making method when executed by a processor.

[0054] The beneficial effects of this invention are as follows: The multi-intelligent node-driven cluster decision-making method provided by this invention achieves dynamic adaptive scheduling of task allocation by constructing a weighted task scoring mechanism based on task complexity coefficients and empirical factors, thereby improving the rationality of task scheduling and the flexibility of system response. By introducing a decentralized master control failure detection and redundant master election mechanism, the fault tolerance and operational stability of the system are effectively improved. Furthermore, the consensus structure dynamic revision process driven by objection feedback enhances the consistency and execution efficiency of task allocation. This invention achieves better results in terms of task scheduling accuracy, system collaboration robustness, and autonomous collaboration capabilities among multiple nodes. Attached Figure Description

[0055] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0056] Figure 1 The following is an overall flowchart of a multi-intelligent node-driven cluster decision-making method provided in the first embodiment of the present invention.

[0057] Figure 2 The system flowchart of a multi-intelligent node driven cluster decision-making method provided in the second embodiment of the present invention is shown. Detailed Implementation

[0058] To make the above-mentioned objects, features, and advantages of the present invention more apparent and understandable, specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the protection scope of the present invention.

[0059] Example 1, referring to Figure 1 As an embodiment of the present invention, a multi-intelligent node-driven cluster decision-making method is provided, comprising:

[0060] S1: Broadcast the node's running parameters in the initial state and elect a master node, generating a dataset of node task awareness results based on the master node.

[0061] After startup, the node enters broadcast mode and sends an initialization broadcast packet within a random delay of 2 seconds.

[0062] The initial broadcast packet includes the node number, battery percentage, communication quality (Qi), and current task buffer information.

[0063] Once a node receives at least a set number of broadcast packets, it enters the RAFT lightweight master election process. While retaining the master election phase, the log replication process is omitted, and nodes with task complexity scores below a preset threshold and guaranteed to have communication quality indicators above the average value are given priority as candidate masters.

[0064] One preferred scheme for selecting candidate master controllers is:

[0065]

[0066] Among them, R i This represents the overall score of the master control candidate for node i. Q represents the task buffer complexity index for node i. i H represents the communication quality score of node i. i E represents the empirical factor of node i. i λ1, λ2, λ3, and λ4 represent the remaining battery percentage of node i, and λ1, λ2, λ3, and λ4 represent the scoring weight parameters.

[0067] By broadcasting local task status awareness information packets from child nodes to the master node, the master node gathers task priority, execution time, remaining energy, and historical completion rate data to form a dataset based on the node task awareness results.

[0068] S2: Calculate the task complexity coefficient and node task experience weight factor based on the dataset of node task perception results, execute the state convergence of the task master node and generate a multi-task consensus decision vector with weighted voting mechanism.

[0069] Each child node generates a task complexity coefficient based on key real-time parameters of the dataset. The weight coefficients are adjusted based on dynamic environmental experience, and the sum of the weights is 1.

[0070] A preferred approach to generating a task complexity coefficient is as follows:

[0071] T ci =α·P i +β·D i +γ·(1-E i )

[0072] Among them, T ci P represents the task complexity coefficient of node i. i Indicates the priority of the task objective, D i E represents the estimated time for task execution. i α represents the remaining battery percentage, and β and γ represent the task complexity weights.

[0073] The complexity coefficient is calculated by combining the weighted sum of the target priority and the estimated execution time with the inverse ratio of the remaining power.

[0074] The completion status of each node in the recent task execution process is statistically analyzed, and the task completion rate of the node is calculated. The completion rate is used as the experience weight factor of the node, and the stability and reliability of the node's task execution are quantified through the experience weight.

[0075] A preferred approach to improving the task completion rate of computing nodes is as follows:

[0076]

[0077] Among them, H i The empirical factor representing node i, This indicates the number of tasks that the node has successfully completed. This indicates the total number of tasks received by the node.

[0078] The task complexity coefficient and empirical weight factor are uploaded to the master node through a state broadcasting mechanism to achieve state convergence of the task master node.

[0079] Once the master node receives the task complexity coefficients and empirical factors reported by all child nodes, it organizes the status information of all nodes into a unified data structure and caches the fields in matrix form.

[0080] The fields cached include the node's unique ID, node task complexity coefficient, node experience weight factor, node communication quality score, and node task load.

[0081] The master node ranks and scores all candidate nodes corresponding to each scheduled task according to a unified scheduling and scoring logic. The scoring criteria are constructed using the inverse relationship between the node's experience factor and the task complexity coefficient.

[0082] One preferred scheme for the master node to score candidate nodes is:

[0083]

[0084] Among them, S i Let represent the overall task score of node i, and ∈ represent a minimal constant.

[0085] The master node selects the node with the highest score to execute the task first, and generates a task number, assigns a node number, and assigns a scheduling score as the task scheduling scheme.

[0086] The task scheduling scheme is integrated into a set of weighted task consensus vectors, which are marked with version numbers by the master node and broadcast to all nodes to enter the consensus confirmation phase.

[0087] During the consensus confirmation process, if node feedback indicates disagreements regarding some task allocations, the master node will dynamically reconstruct the consensus vector based on the feedback suggestions and rebroadcast it until a stable and consistent task allocation structure is formed.

[0088] After the master node completes the task status convergence, it constructs a multi-task weighted scheduling and scoring model based on the task complexity coefficients and task completion history factors uploaded by each child node.

[0089] The master node calculates the voting weight score of the node in the current scheduling round based on the task complexity coefficient and task completion history rate of the child node.

[0090] The voting weight score is determined by the inverse relationship between the node's experience factor and the task complexity coefficient. The higher the node's score, the greater its weight in the voting mechanism.

[0091] For each task to be scheduled, the master node selects the nodes with higher voting weight scores from all candidate nodes of executable tasks as the candidate set.

[0092] The master node normalizes the voting weights of each node in the candidate set and constructs a weight vector structure for task allocation.

[0093] A preferred approach to normalize the voting weights of each node in the candidate set is as follows:

[0094]

[0095] Among them, w ij S represents the normalized voting weight of node i for task j. ij This represents the score of node i for task j, n j This represents the number of candidate nodes for task j.

[0096] The vector structure consists of triplets, namely, task number, candidate node number, and weighted score value of the corresponding node.

[0097] The master node prioritizes the scheduling node for tasks based on the weighted score value, and combines the task number with the optimal node number to generate a preliminary task consensus decision vector.

[0098] A preferred approach for generating the initial task consensus decision vector is as follows:

[0099]

[0100] Among them, C t Let T represent the multi-task consensus decision vector for the t-th round of scheduling. j ID represents the j-th task number. ij w represents the node number of the selected task j. ij This represents the weighted score of the node for task j, and m represents the number of tasks in this round.

[0101] The task consensus decision vector is broadcast to all child nodes in the form of a structured data packet, and a consistency confirmation window is opened. Each child node returns confirmation and objection feedback flags based on the received allocation results and its own task awareness.

[0102] A preferred approach for returning confirmation and objection feedback indicators is:

[0103]

[0104] Among them, F consensus N represents the consensus feedback consistency rate of this round of tasks. agree N represents the number of nodes that have not yet confirmed the feedback. total This indicates the total number of nodes participating in the feedback.

[0105] The master node adjusts the weights of conflicting task assignments based on the feedback data returned by the child nodes. It uses a feedback frequency-weighted reconstruction method to dynamically correct the task weight values ​​of the corresponding nodes and forms a revised consensus vector marked with a version number.

[0106] A preferred approach for generating the revised consensus vector is:

[0107]

[0108] in, This indicates the revised task score. Indicates the original score. Let δ represent the average frequency of feedback conflicts at node i, and let δ represent the score adjustment coefficient.

[0109] When the consensus feedback rate reaches a preset threshold, the master node solidifies the current consensus decision vector version as the task allocation and execution benchmark within the scheduling cycle, and distributes it to all nodes for synchronous execution.

[0110] S3: After detecting the failure of the master node, the redundant master election process is initiated through the preset heartbeat judgment mechanism and the consensus update under the new master is completed.

[0111] All non-master nodes are configured with a fixed master heartbeat listening period, with a default value of 10 seconds. During each listening period, they attempt to receive the heartbeat signal broadcast by the master node. The heartbeat signal is a predefined short status synchronization packet, which includes the master node number, task consensus version number, broadcast timestamp, and master running status summary.

[0112] Each child node is equipped with a packet loss counter. If it fails to receive a heartbeat packet from the master node for three consecutive heartbeat cycles, it will automatically mark the master node as suspected of failure and enter the local judgment waiting stage.

[0113] A preferred scheme for counting lost heartbeats is:

[0114]

[0115] Among them, L i This represents the number of times node i has not received a master control packet in the last K heartbeats. This indicates whether the master heartbeat was successfully received in the Kth heartbeat cycle, and 1(·) represents the indicator function.

[0116] During the local determination phase, the node will attempt to send a master control activity confirmation packet to the master control node via unicast. If there is still no response to the confirmation packet within three consecutive heartbeat cycles, the master control status will be officially set to unavailable, and a master control failure notification message will be broadcast in the network.

[0117] A preferred method for determining the master control status is:

[0118] Version t+1 =Version t +1

[0119] Among them, Version t Indicates the current consensus task scheduling version number, Version t+1 This indicates the consensus version number for the new round of tasks.

[0120] Upon receiving the master controller failure notification, all nodes immediately cease their original consensus scheduling operations and uniformly enter the candidate master controller information broadcasting phase, where each node broadcasts its own master controller candidate status packet.

[0121] The broadcast's own master control candidate status packet includes the node number, the current task cache quantity and total complexity index, the remaining power percentage, the current communication quality index, and the node's historical task completion rate.

[0122] After broadcasting, all nodes wait to receive candidate information from other nodes. After the statistics are completed, each node will independently calculate the score based on a unified comprehensive scoring function. The node with the highest score waits for the master control confirmation packet to be broadcast after the broadcast round ends and requests master control confirmation feedback from the entire network.

[0123] The broadcast master confirmation packet includes its own score, version number, and candidate proof packet summary.

[0124] If there are no objections after receiving the response, a confirmation response is returned. When the node responses are consistent, the node officially switches to the new master state. The new master node inherits the consensus task cache structure of the previous master and immediately starts the state recovery and task reconstruction process.

[0125] After the new master node confirms the takeover, it first loads the latest task consensus vector and version number from the original master broadcast cache, and performs consistency checks on the status of all incomplete tasks.

[0126] If the node feedback consistency ratio exceeds the set threshold, the new master node confirms the validity of this round of revision consensus, officially enters the task scheduling and synchronous execution process, and increments the consensus version number by 1, recording it as the current task version.

[0127] If no consensus is reached, the master node will perform a single secondary revision based on the conflicting task number, redistribute tasks within a limited scope, and conduct up to N rounds of feedback iteration. If no effective consensus is still formed, the master node will use a forced instruction method to complete the task assignment and mark the downgraded consensus version to enter the downgraded operation state.

[0128] Furthermore, the new master node queries the current status of the execution node corresponding to each task objective based on the consensus task allocation record, and determines whether the node is online normally and whether there are any abnormal situations such as heartbeat interruption, task response timeout or insufficient power.

[0129] For task execution nodes that are determined to be in failure or abnormal status, the new master node marks their task status as "mismatched" and extracts the original target, parameters, and candidate node pool information of the task from the task log.

[0130] The new master node reconstructs the scheduling candidate set of tasks to be repaired based on the latest node status table, re-evaluates the task complexity coefficient and experience score, executes the consensus voting reconstruction process, and forms a new round of task allocation results.

[0131] The result was reorganized into a weighted voting consensus vector structure, which includes task number, candidate node number, refactoring score, current version number, and revision identifier.

[0132] After the new consensus vector is constructed, the new master node broadcasts a "consensus revision notice" to all child nodes and starts the task consensus confirmation window. Each node returns a confirmation or rejection response flag based on its local state.

[0133] Example 2, refer to Figure 2 As an embodiment of the present invention, a multi-intelligent node driven cluster decision-making system is provided, including a data acquisition and generation module 100, a decision calculation module 200, and a detection and update module 300.

[0134] S4: The acquisition and generation module 100 is used to broadcast the node running parameters in the node initialization state and to elect the master node, and to generate a dataset of node task perception results based on the master node.

[0135] It should also be noted that during the initialization of each intelligent node, the data acquisition and generation module 100 is responsible for broadcasting the node's operating parameters, including node number, power information, communication quality indicators, and task buffer status, and initiating a lightweight master node election process during the broadcast. After the election is completed, each child node sends local task perception information to the selected master node, and the master node constructs a dataset of node task perception results based on this information and synchronously transmits the dataset to the decision computing module 200.

[0136] S5: The decision calculation module 200 is used to calculate the task complexity coefficient and node task experience weight factor based on the dataset of node task perception results, execute the state convergence of the task master node, and generate a multi-task consensus decision vector with a weighted voting mechanism.

[0137] It should also be noted that the decision calculation module 200 receives and parses the above-mentioned node task-aware dataset, performs dynamic calculation of the task complexity coefficient for each node and generates empirical factors based on historical statistics of task completion rate; subsequently, the module 200 performs weighted scheduling scoring and ranking of candidate nodes according to a unified scoring model, constructs a multi-task consensus decision vector and broadcasts it, and enters the consistency feedback confirmation stage; during the feedback process, the module 200 continuously monitors the consensus feedback response and dynamically revises and updates the vector for task allocation items with disagreements.

[0138] S6: The detection and update module 300 is used to initiate a redundant master election process and complete the consensus update under the new master after detecting the failure of the master node through a preset heartbeat judgment mechanism.

[0139] It should also be noted that the detection and update module 300 operates in a guardian mode, continuously listening to the heartbeat broadcast of the master node during task execution. Once the heartbeat of the master node is abnormal and exceeds the preset periodic threshold, module 300 initiates the redundant master election mechanism and recalculates the master candidate score value based on the status of the standby node. After the new master node is established, module 300 is responsible for loading the old consensus task cache, performing consistency verification and reconstruction of unfinished tasks, and rebroadcasting the new task decision vector to all child nodes through the consensus update mechanism to ensure the continuity and stability of the system in the event of a master node change.

[0140] If a function is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0141] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as a sequenced list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by, or in conjunction with, an instruction execution system, apparatus, or device (such as a computer-based system, a processor-included system, or other system that can fetch and execute instructions from, an instruction execution system, apparatus, or device). For the purposes of this specification, "computer-readable medium" can be any means that can contain, store, communicate, propagate, or transmit programs for use by, or in conjunction with, an instruction execution system, apparatus, or device.

[0142] More specific examples (a non-exhaustive list) of computer-readable media include: electrical connections (electronic devices) having one or more wires, portable computer disk drives (magnetic devices), random access memory (RAM), read-only memory (ROM), erasable and editable read-only memory (EPROM or flash memory), fiber optic devices, and portable optical disc read-only memory (CDROM). Furthermore, computer-readable media can even be paper or other suitable media on which programs can be printed, because programs can be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, interpreting, or otherwise processing as necessary, and then stored in computer memory.

[0143] It should be understood that various parts of the present invention can be implemented using hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented using software or firmware stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware, as in another embodiment, it can be implemented using any one or a combination of the following techniques known in the art: discrete logic circuits having logic gates for implementing logical functions on data signals, application-specific integrated circuits (ASICs) having suitable combinational logic gates, programmable gate arrays (PGAs), field-programmable gate arrays (FPGAs), etc. It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A multi-intelligent node driven cluster decision method, characterized in that, The application relates to a node running parameter initialization state of a broadcast node and a master node election, a node task perception result data set generated by the master node, a task complexity coefficient and a node task experience weight factor calculated based on the node task perception result data set, a state convergence of a task master node and a multi-task consensus decision vector with a weighted voting mechanism generated, a redundancy master election process started through a preset heartbeat judgment mechanism and a consensus update under a new master completed after a master node failure is detected, a task complexity coefficient and a node task experience weight factor calculated, a task complexity coefficient dynamically calculated by a node based on a task priority, an execution time length and a residual energy, a node task experience weight factor generated in combination with a historical completion rate, a score sorting and a scheduling allocation performed after the master node is uploaded, the master node including a consensus vector constructed by the master node using a voting weight normalization strategy, a dissent feedback correction mechanism introduced, a dynamic reconstruction and a version management of a task consensus structure performed, the master node election including a broadcast state entered by a node after being started, a random delay sending initialization broadcast packet set, the initialization broadcast packet including a node number, a power percentage, a communication quality Qi and current task buffer information, a RAFT lightweight master election process entered by a node after receiving no less than a set number of broadcast packets, a log replication process omitted on the premise of a master selection stage reserved, a node with a task complexity score lower than a preset threshold and a communication quality index higher than an average value selected as a candidate master being preferentially selected, the multi-task consensus decision vector with the weighted voting mechanism generated including a multi-task weighted scheduling score model constructed by the master node based on the task complexity coefficient and the task completion history factor uploaded by each subnode after a task state convergence of the master node is completed, a voting weight score of a node in a current scheduling round calculated by the master node according to the task complexity coefficient and the task completion history rate of the subnode, the voting weight score determined by a reverse relationship between a node experience factor and the task complexity coefficient, a higher score of the node, a higher weight in the voting mechanism, a candidate set of nodes selected from all executable task candidate nodes of each task to be scheduled by the master node, a weight vector structure of task allocation constructed by the master node after a voting weight normalization processing of each node in the candidate set, the vector structure composed of three tuples, respectively a task number, a candidate node number and a weighted score value of a corresponding node, the redundancy master election process started through the preset heartbeat judgment mechanism including a fixed master heartbeat listening period set for all non-master nodes, a default value being 10 seconds, a heartbeat signal broadcasted by the master node tried to be received in each listening period, the heartbeat signal being a predefined short state synchronization packet, including a master node number, a task consensus version number, a broadcast timestamp and a master running state digest, a packet loss counter set for each subnode, the master state marked as a failure suspicious and a local judgment waiting stage entered when the subnode fails to successfully receive the heartbeat packet of the master node for three continuous heartbeat periods. ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ ​ In the local decision phase, the node will try to send a master active confirmation packet to the master node by unicast, and if the master node is still unresponsive within the set three consecutive heartbeat periods, the master state will be officially set as unavailable, and a master failure notification message will be broadcast in the network; After receiving the master failure notification, all nodes immediately stop the original consensus scheduling operation and enter the candidate master information broadcast phase. Each node will broadcast its master candidate state packet; Broadcasting its master candidate state packet includes node number, current task cache quantity and total complexity index, remaining battery percentage, current communication quality index, and node task historical completion rate; After broadcasting, all nodes wait to receive the candidate information of other nodes. After the statistics are completed, each node will independently calculate the score based on the unified comprehensive scoring function. The node with the highest score will broadcast a master confirmation packet after the end of the broadcast round and request the entire network to perform master confirmation feedback; The broadcast master confirmation packet includes its own score, version number, and candidate proof packet digest; After receiving no objections, return the confirmation response. When the node responses are consistent, the new node master state is officially switched. The new master node inherits the consensus task cache structure of the previous master and immediately starts the state recovery and task reconstruction process; The completion of the consensus update under the new master includes, After the new master node confirms the succession, it loads the latest task consensus vector and version number from the original master broadcast cache and performs consistency verification on all unfinished task states; If the consistency feedback rate exceeds the set threshold, the new master node confirms that the revised consensus is valid for this round, officially enters the task scheduling synchronous execution process, and increments the consensus version number by 1 to record the current task version; If no consensus is reached, the master node performs single secondary revision on the conflicting task number based on the feedback, reassigns within the limited range, and performs up to N rounds of feedback iteration. If an effective consensus is still not formed, the master node completes the task assignment using the master forced instruction method and marks the degraded consensus version to enter the degraded running state.

2. The multi-intelligent node driven swarm decision method of claim 1, wherein: The data set generated by the node task perception result includes, The master node converges the task priority, execution time, remaining energy, and historical completion rate data by broadcasting local task state perception information packets to the master node, forming a data set based on the node task perception result.

3. The multi-intelligent node driven swarm decision method of claim 1, wherein: The calculation of the task complexity coefficient and the node task experience weight factor includes, Each sub-node generates a task complexity coefficient based on the key real-time parameters of the data set, and the weight coefficient is adjusted dynamically according to the environment; The complex task degree coefficient includes the weighted sum of the target priority and the estimated execution time, combined with the reciprocal proportion of the remaining battery capacity to calculate the complex task degree coefficient; Statistical data on the completion of each node in the recent task execution process is collected to calculate the node's task completion rate, which is used as the node's experience weight factor to quantify the stability and reliability of the node's task execution; The task complexity coefficient and the experience weight factor are uploaded to the master node through the state broadcast mechanism for state convergence of the task master node.

4. The multi-intelligent node driven swarm decision method of claim 1, 2 or 3, wherein: The state convergence of the task master node includes, When the master node receives the task complexity coefficients and experience factors reported by all sub-nodes, the state information of all nodes is organized into a unified data structure, and the fields are cached in a matrix form; The cached fields include node unique number, node task complexity coefficient, node experience weight factor, node communication quality score, and node task load condition; The master node scores and sorts all candidate nodes corresponding to each task to be scheduled according to a unified scheduling score logic, and the scoring standard is constructed by using the inverse ratio relationship between the experience factor and the task complexity coefficient of the node; The master node selects the node with the highest score to execute the task first, and generates a task number, assigns a node number, and schedules a score as a task scheduling scheme; The task scheduling scheme is integrated into a set of weighted task consensus vectors, which are marked with a version number by the master node and distributed to all nodes by broadcasting, and enter the consensus confirmation stage; During the consensus confirmation process, the node feedback result shows that there is an objection to part of the task allocation, and the master node will dynamically reconstruct the consensus vector according to the feedback suggestion and broadcast it again until a stable and consistent task allocation structure is formed.

5. The multi-intelligent node driven swarm decision method of claim 4, wherein: The generation of the multi-task consensus decision vector with the weighted voting mechanism further includes, The master node determines the scheduling node of the task in priority according to the weighted score value, and combines the task number with the optimal node number to generate a preliminary task consensus decision vector; The task consensus decision vector is broadcast to all sub-nodes in a structured data packet form, and a consistency confirmation window is opened, each sub-node returns a confirmation and objection feedback identifier according to the received allocation result and its own task perception; The master node adjusts the weight of the task allocation with conflicts according to the feedback data returned by the sub-nodes, dynamically corrects the task weight value of the corresponding node by using the feedback frequency weighted reconstruction method, and forms a revised version of the consensus vector in the form of version number marking; When the consensus feedback rate reaches the preset threshold, the master node solidifies the current consensus decision vector version as the execution benchmark of the task allocation in the scheduling period, and issues it to all nodes for synchronous execution.

6. A multi-intelligent node driven swarm decision system, characterized by: It includes a collection and generation module (100), a decision calculation module (200), and a detection update module (300); The collection and generation module (100) is used for broadcasting the node running parameters in the initialization state of the node and performing master node election, and generating a data set of node task perception results according to the master node; The decision calculation module (200) is used for calculating the task complexity coefficient and the node task experience weight factor based on the data set of the node task perception results, performing state convergence of the task master node, and generating a multi-task consensus decision vector with a weighted voting mechanism; The detection update module (300) is used for starting a redundancy master election process and completing consensus update under a new master node through a preset heartbeat judgment mechanism when the master node fails; The task complexity coefficient and the node task experience weight factor include that the node dynamically calculates the task complexity coefficient based on the task priority, the execution time length and the residual energy, and generates the node task experience weight factor in combination with the historical completion rate, uploads the master node, and then performs scoring, sorting and scheduling allocation; The master node includes that the master node adopts a voting weight normalization strategy to construct a consensus vector, introduces a dissent feedback correction mechanism, and performs dynamic reconstruction and version management of the task consensus structure; The master node election includes, After starting, the node enters a broadcast state, and sets a random delay to send an initialization broadcast packet; The initialization broadcast packet includes a node number, a power percentage, a communication quality Qi and current task buffer information; When the node receives no less than a set number of broadcast packets, it enters a RAFT lightweight master election process, omits the log replication process under the premise of retaining the master election stage, and preferentially selects a node with a task complexity score lower than a preset threshold and ensuring that the communication quality index is higher than the average value as a candidate master; The generation of the multi-task consensus decision vector with a weighted voting mechanism includes, After the master node completes the task state convergence, a multi-task weighted scheduling score model is constructed based on the task complexity coefficient and the task completion history factor uploaded by each sub-node; The master node calculates the voting weight score of the node in the current scheduling round according to the task complexity coefficient and the task completion history rate of the sub-node; The voting weight score is determined by the inverse relationship between the node experience factor and the task complexity coefficient, and the higher the node score, the greater the weight in the voting mechanism; The master node selects the nodes with higher voting weight scores from all executable task candidate nodes as a candidate set for each task to be scheduled; The master node normalizes the voting weight of each node in the candidate set to construct a weight vector structure for task allocation; The vector structure is composed of three tuples, namely task number, candidate node number and weighted score value of the corresponding node; The redundant master election process started through the preset heartbeat judgment mechanism includes, All non-master nodes set a fixed master heartbeat listening period, with a default value of 10 seconds, and attempt to receive the heartbeat signal broadcast by the master node in each listening period. The heartbeat signal is a pre-defined short state synchronization packet, which includes the master node number, the task consensus version number, the broadcast timestamp and the master running state summary; Each sub-node is provided with a packet loss counter. When it fails to successfully receive the heartbeat packet of the master node for three consecutive heartbeat periods, it automatically marks the master state as invalid and suspicious, and enters a local judgment waiting stage; In the local judgment stage, the node will attempt to send a master activity confirmation packet to the master node through unicast. If the confirmation packet still has no response within a set of three consecutive heartbeat periods, the master state is officially set as unavailable, and a master failure notification message is broadcast in the network; After receiving the master failure notification, all nodes immediately stop the original consensus scheduling operation and uniformly enter the candidate master information broadcast stage, and each node broadcasts its master candidate state packet. The broadcasted master candidate state package includes a node number, a current task cache number, a total complexity index, a remaining battery percentage, a current communication quality index, and a node task history completion rate; After the broadcast, all nodes wait to receive candidate information from other nodes. After the statistics are completed, each node independently calculates a score result based on a unified comprehensive scoring function. The node with the highest score waits for a master confirmation package after the end of the broadcast round and requests the entire network to perform master confirmation feedback; The broadcast master confirmation package includes a score, a version number, and a candidate proof package digest; After receiving no objections, the node returns a confirmation response. When the node responses are consistent, the node is officially switched to a new node master state. The new master node inherits the consensus task cache structure of the previous master and immediately starts the state recovery and task reconstruction process; The consensus update under the new master includes, After the new master node confirms the succession, the latest task consensus vector and version number are loaded from the broadcast cache of the original master, and consistency verification is performed on all unfinished task states; When the consistency feedback proportion exceeds a set threshold, the new master node confirms that the revised consensus of this round is valid, officially enters the task scheduling synchronization execution process, and sets the consensus version number +1 as the current task version. When no consensus is reached, the master node performs single secondary revision according to the feedback conflict task number, reassigns within a limited range, and performs at most N rounds of feedback iteration. If an effective consensus is still not formed, the master node completes task assignment using a master forced instruction method and marks the degraded consensus version to enter a degraded running state. 7.A computer device, comprising a memory and a processor, wherein the memory stores a computer program, and the computer device is configured to perform the method according to any one of claims 1-6 when the computer program is executed by the processor. The processor executes the computer program to implement the steps of the multi-intelligent node-driven cluster decision method of any one of claims 1 to 5.

8. A computer-readable storage medium having stored thereon a computer program, characterized in that, The computer program is executed by the processor to implement the steps of the multi-intelligent node-driven cluster decision method of any one of claims 1 to 5.

Images