WO2025232168A1 - 分布式启发的无人集群类脑群智融合搜索围捕方法 - Google Patents

分布式启发的无人集群类脑群智融合搜索围捕方法

Info

Publication number
WO2025232168A1
WO2025232168A1 PCT/CN2024/136298 CN2024136298W WO2025232168A1 WO 2025232168 A1 WO2025232168 A1 WO 2025232168A1 CN 2024136298 W CN2024136298 W CN 2024136298W WO 2025232168 A1 WO2025232168 A1 WO 2025232168A1
Authority
WO
WIPO (PCT)
Prior art keywords
target
agent
search
space
det
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Pending
Application number
PCT/CN2024/136298
Other languages
English (en)
French (fr)
Inventor
张晖
张豪杰
唐安安
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Nanjing University of Posts and Telecommunications
Original Assignee
Nanjing University of Posts and Telecommunications
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Nanjing University of Posts and Telecommunications filed Critical Nanjing University of Posts and Telecommunications
Publication of WO2025232168A1 publication Critical patent/WO2025232168A1/zh
Pending legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Classifications

    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/20—Analysis of motion
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/004—Artificial life, i.e. computing arrangements simulating life
    • G06N3/006—Artificial life, i.e. computing arrangements simulating life based on simulated virtual individual or collective life forms, e.g. social simulations or particle swarm optimisation [PSO]
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N5/00—Computing arrangements using knowledge-based models
    • G06N5/01—Dynamic search techniques; Heuristics; Dynamic trees; Branch-and-bound
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N5/00—Computing arrangements using knowledge-based models
    • G06N5/04—Inference or reasoning models
    • G06N5/043—Distributed expert systems; Blackboards
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/20—Image preprocessing
    • G06V10/25—Determination of region of interest [ROI] or a volume of interest [VOI]
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
    • G06V10/77—Processing image or video features in feature spaces; using data integration or data reduction, e.g. principal component analysis [PCA] or independent component analysis [ICA] or self-organising maps [SOM]; Blind source separation
    • G06V10/80—Fusion, i.e. combining data from various sources at the sensor level, preprocessing level, feature extraction level or classification level
    • G06V10/809—Fusion, i.e. combining data from various sources at the sensor level, preprocessing level, feature extraction level or classification level of classification results, e.g. where the classifiers operate on the same input data
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/20—Special algorithmic details
    • G06T2207/20081—Training; Learning
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/20—Special algorithmic details
    • G06T2207/20084—Artificial neural networks [ANN]
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/30—Subject of image; Context of image processing
    • G06T2207/30241—Trajectory
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V2201/00—Indexing scheme relating to image or video recognition or understanding
    • G06V2201/07—Target detection
    • Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
    • Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
    • Y02T—CLIMATE CHANGE MITIGATION TECHNOLOGIES RELATED TO TRANSPORTATION
    • Y02T10/00—Road transport of goods or passengers
    • Y02T10/10—Internal combustion engine [ICE] based vehicles
    • Y02T10/40—Engine management systems

Definitions

  • This invention relates to a distributed, heuristic-inspired, unmanned swarm brain-like swarm intelligence fusion search and capture method, and more particularly to a swarm intelligence decision-making technology using heuristic optimization algorithms, belonging to the field of artificial intelligence and swarm intelligence technology.
  • Distributed-inspired swarm intelligence fusion technology is one of the main methods for multi-agent collaborative search and capture.
  • unmanned combat scenarios it has the characteristics of wide coverage, high difficulty in search and capture, and difficulty in target location.
  • problems such as motion ambiguity are easily generated, resulting in low target detection efficiency.
  • the purpose of this invention is to address the shortcomings and deficiencies of existing technologies by proposing a distributed-inspired, brain-like swarm intelligence fusion search and encirclement method for unmanned swarms.
  • First based on YOLOv5s target detection technology, it addresses the issue of real-time target movement by using predicted trajectories to track dynamic targets.
  • a decision-level brain-like fusion mechanism is then used to complete the decision fusion of multiple agents, reducing redundant region searches.
  • a multi-agent information-level brain-like fusion encirclement mechanism is proposed based on the role update mechanism of the Grey Wolf optimization algorithm, combined with dynamic Voronoi region partitioning. This mechanism accurately updates the encirclement endpoints of agents, completing the task of encircling moving targets.
  • the technical solution adopted by this invention to solve its technical problem is: a distributed-inspired unmanned cluster brain-like swarm intelligence fusion search and capture method, which includes the following steps:
  • Step 1 Perform single target detection based on the Yolov5 network, and combine it with the DeepSort algorithm to predict the possible location of the target in the next frame. Calculate the normalized distance difference between the point in the predicted box and the central axis, and the normalized area difference between the predicted box area and the defined imaging area, denoted as diff det .
  • Step 2 Map the coordinates of the agent's navigation endpoint in physical space based on the distance difference and area difference.
  • Step 3 Construct a chaotic mapping algorithm that coordinates diffusion and contraction to map the update of the navigation endpoint coordinates to a uniformly distributed chaotic space.
  • Step 4 Based on the target detection confidence decision information and region location information of all agents, fuse them to generate the search space.
  • Global search information Based on global search information, the dynamic decision-making agent performs both expansion and contraction search tasks.
  • Step 5 Based on the global search information of the agents, integrate various sensor data from the multi-agent system, estimate the target location information, divide the encirclement area based on the dynamic Voronoi diagram, and accurately determine the encirclement position of the agents.
  • Step 6 Based on the agent's real-time position and obstacle positions, and simultaneously based on the target object's V goal map area in the Voronoi diagram, detect the target object's V goal map area. and boundaries, if area Less than the threshold If all boundaries are adjacent to a Voronoi diagram, then the encirclement ends.
  • step 1 single target detection is performed based on the Yolov5 network, and the DeepSort algorithm is used to predict the possible location of the target in the next frame.
  • width img and height img represent the width and height of the image, respectively.
  • step 2 the coordinates of the intelligent agent's navigation endpoint in physical space are mapped based on the distance difference and area difference. Specifically, this includes: first, estimating the distance the agent needs to move forward and the distance it needs to travel to the left and right based on the normalized distance difference and area difference.
  • ⁇ represents the maximum forward and left/right movement distances, respectively; ⁇ is the minimum real number, indicating that the agent stops moving forward and backward when the detection box area is greater than the optimal area, while ensuring the feasibility of subsequent angle calculations; and the mapped navigation endpoint coordinates in the agent's global coordinate system.
  • the real-time coordinates of the intelligent agent in, and The real-time coordinates of the intelligent agent.
  • constructing a diffusion-contraction coordinated chaotic mapping algorithm to map the update of the navigation endpoint coordinates to a uniformly distributed chaotic space specifically includes: firstly, dividing the global space of M space ⁇ M space into m ⁇ m local spaces. When iterating ut times, the global space utM space ⁇ utM space is divided into utm ⁇ utm local spaces. Then, the search is performed outward from the starting point. Therefore, the target position of each agent needs to be estimated using an improved chaotic mapping.
  • the improved chaotic diffusion factor expression is as follows:
  • ⁇ sub> t ⁇ /sub> is a control parameter with a value range of (0,1).
  • S ⁇ sub>i ⁇ /sub> be a chaotic mapping sequence with values ranging from (0,1). This sequence maps the initial search endpoint of the agent within an M ⁇ M space search space, where the origin is located at the midpoint of the M ⁇ M space .
  • the expression for estimating the target position of agent S ⁇ sub> i ⁇ /sub> is:
  • ut represents the number of chaotic mapping iterations.
  • the value of ut is 1. and These are the chaos mapping factors for the x-axis and y-axis, respectively.
  • step 3 includes: if an agent finds a target, i.e., a suspected target space exists, then a chaotic mapping is performed in the local space, mapping the chaotic factors of the x-axis and y-axis, both in the range [-1, 1], to the chaotic factor mapping range of the suspected target space.
  • the mapping range of the chaotic contraction factor is:
  • the search space is generated by fusing the target detection confidence decision information and regional location information of all agents.
  • Global search information includes the following:
  • the decision information of the intelligent agent includes the following:
  • agent Si The target detection confidence of the intelligent agent.
  • the location of agent Si in the search space The location of agent Si in the search space; global search information. Specifically, it is expressed as follows:
  • C wh represents the local space
  • B wh represents the number of agents that searched the local space.
  • the confidence level of target detection within the local space is determined by the local space.
  • the target confidence scores detected by all agents within the system are fused to generate the target. Specifically, it is expressed as follows:
  • the dynamic decision-making agent performs diffusion search and contraction search tasks based on global search information, specifically including: if All are 0, but there are intelligent agents. If True, meaning an agent has reached the search endpoint, then a spatial diffusion search is performed; if... All values are 0, indicating no agent detected the target, but agents are present. If the value is True, it means an agent has reached the search endpoint. In this case, the search endpoints for all agents need to be updated, and the search range needs to be expanded, meaning the number of iterations ut increases. Because the search range expands, the local region... As this changes, it becomes necessary to inherit historical global information.
  • the inheritance formula is as follows:
  • the value is not 0, it indicates that an agent Si has detected the target and is performing a local agent space shrinkage search.
  • Corresponding local space As a suspected target space
  • the agents within eight local spaces around the target estimate the target location, thus achieving a shrinking search.
  • step 5 the estimation of target location information based on the global search information of the intelligent agent, by fusing various sensor data from the multi-agent system and combining the Gray Wolf optimization algorithm, specifically includes the following:
  • the perception information of the intelligent agent is as follows:
  • N is the number of agents; based on the agent's own state vector
  • the target detection probability is used to calculate the individual fitness value. Role assignment is completed based on the ranking of fitness values. The higher the target detection probability of an agent, the higher its fitness, indicating the emergence of a leader.
  • the fitness value calculation formula is as follows:
  • the encirclement area is divided based on a dynamic Voronoi diagram.
  • the precise encirclement position of the agent specifically includes: the estimated target coordinates (x goal , y goal ) and the coordinates of nearby obstacles. Including the coordinates of the alpha wolf, there are a total of seven coordinates.
  • Initializing the Voronoi diagram yields seven discrete regions. and and the vertex set of each region and Calculate the centroid of the target object V in the goal diagram.
  • the formula for calculating the center of gravity is as follows:
  • the area of the discrete Voronoi diagram region V_goal is used as the basis for the wolf detection position update formula in the gray wolf optimization algorithm. Then, the centroid position of the target region is used to improve the wolf detection position update formula, thereby generating the updated wolf detection position.
  • the improved formula is as follows:
  • r1 and r2 are random vectors with values between [0,1], and a is a control parameter that decreases linearly from 2 to 0 during the iteration process.
  • step 6 includes: based on the real-time position of the agent and the position of the obstacle, if an agent detects a target, updating the agent's role using the Gray Wolf Optimization Algorithm, and updating the coordinates of obstacles near the leader wolf. and and target coordinates Generate a new Voronoi diagram; if no agent detects the target, but the target is moving, update the agent's decision and the coordinates of the obstacle after a time T of movement. and and target coordinates Generate a new Voronoi diagram; simultaneously, detect the area and boundary of the target object's Voronoi diagram, if the area... Less than the threshold If all boundaries are adjacent to a Voronoi diagram, then the encirclement ends.
  • This invention is based on YOLOv5s target detection technology. To address the problem of real-time target motion, it uses predicted trajectories to complete the tracking task of dynamic targets.
  • This invention addresses the problem of large search areas in unmanned combat scenarios by improving the chaotic mapping algorithm based on the principle of regional diffusion and contraction. It maps the search endpoint of unmanned intelligent agents in a coordinated diffusion and contraction manner and utilizes a decision-level brain-like fusion mechanism to complete the decision fusion of multiple intelligent agents, thereby reducing repeated regional searches.
  • this invention proposes a multi-agent information-level brain-like fusion encirclement mechanism based on the role update mechanism of the gray wolf optimization algorithm and combined with dynamic Voronoi region partitioning. This mechanism accurately updates the encirclement endpoints of the agents, thereby completing the task of encircling the moving target.
  • Figure 1 is a flowchart of the distributed-inspired unmanned cluster brain-like swarm intelligence fusion search and capture method of the present invention.
  • Figure 2 is a layout diagram of the target detection box and the image of the present invention.
  • Figure 3 shows the search range as the number of iterations increases in this invention.
  • Figure 4 shows the end-of-encirclement marker of the present invention.
  • this invention proposes a distributed-inspired unmanned cluster brain-like swarm intelligence fusion search and capture method, which includes the following steps:
  • Step 1 Perform single target detection based on the Yolov5 network, and combine the DeepSort algorithm to predict the position of the target in the next frame. Calculate the normalized distance difference between the point in the predicted box and the central axis, and the normalized area difference between the predicted box area and the limited imaging area, denoted as diff det .
  • Step 2 Map the coordinates of the agent's navigation endpoint in physical space based on the distance difference and area difference.
  • Step 3 Construct a chaotic mapping algorithm that coordinates diffusion and contraction to map the update of the navigation endpoint coordinates to a uniformly distributed chaotic space;
  • Step 4 Based on the target detection confidence decision information and region location information of all agents, fuse them to generate the search space.
  • Global search information Based on global search information, the dynamic decision-making agent performs both expansion search and contraction search tasks.
  • Step 5 Based on the global search information of the agents, integrate various sensor data from the multi-agent system, estimate the target location information, divide the encirclement area based on the dynamic Voronoi diagram, and accurately determine the encirclement position of the agents.
  • Step 6 Based on the agent's real-time position and obstacle positions, and simultaneously based on the target object's V goal map area in the Voronoi diagram, detect the target object's V goal map area. and boundaries, if area Less than the threshold If all boundaries are adjacent to a Voronoi diagram, then the encirclement ends.
  • Step 1 of the present invention includes: performing single target detection based on the Yolov5 network, and combining it with the DeepSort algorithm to predict the possible location of the target in the next frame.
  • width img and height img are the width and height of the image, respectively, and the layout of the detection box and the image is shown in Figure 2.
  • Step 2 of the present invention includes: performing single target detection based on a Yolov5 network, and mapping the navigation endpoint coordinates of the agent in physical space based on distance difference and area difference. First, based on the normalized distance difference and area difference, estimate the distance the agent needs to move forward and the distance it needs to travel to the left and right:
  • ⁇ represents the maximum forward and left/right movement distances, respectively; ⁇ is the minimum real number, indicating that the agent stops moving forward and backward when the detection box area is greater than the optimal area, while ensuring the feasibility of subsequent angle calculations; and the mapped navigation endpoint coordinates in the agent's global coordinate system.
  • the real-time coordinates of the intelligent agent in, and The real-time coordinates of the intelligent agent.
  • Step 3 of the present invention includes: constructing a diffusion-contraction coordinated chaotic mapping algorithm to map the update of the navigation endpoint coordinates to a uniformly distributed chaotic space, firstly dividing the global space of M space ⁇ M space into m ⁇ m local spaces. When iterating ut times, the global space of utM space ⁇ utM space is divided into utm ⁇ utm local spaces. The iterative update space is shown in Figure 3. Then, the search is performed outward from the starting point. Therefore, an improved chaotic mapping is needed to estimate the target position of each agent.
  • the expression for the improved chaotic diffusion factor is as follows:
  • ⁇ sub> t ⁇ /sub> is a control parameter with a value range of (0,1).
  • S ⁇ sub>i ⁇ /sub> be a chaotic mapping sequence with values ranging from (0,1). This sequence maps the initial search endpoint of the agent within an M ⁇ M space search space, where the origin is located at the midpoint of the M ⁇ M space .
  • the expression for estimating the target position of agent S ⁇ sub> i ⁇ /sub> is:
  • step 3 of this invention If an agent finds a target in step 3 of this invention, i.e., a suspected target space exists, then a chaotic mapping is performed in the local space.
  • the chaotic factors of the x-axis and y-axis, both ranging from [-1, 1], are mapped to the chaotic factor mapping range of the suspected target space.
  • the mapping range of the chaotic contraction factor is:
  • Step 4 of the present invention includes: fusing the target detection confidence decision information and regional location information of all agents to generate a search space.
  • Global search information The decision information of the intelligent agent includes the following:
  • agent Si The target detection confidence of the intelligent agent.
  • the location of agent Si in the search space The location of agent Si in the search space; global search information. Specifically, it is expressed as follows:
  • C wh represents the local space
  • B wh represents the number of agents that searched the local space.
  • the confidence level of target detection within the local space is determined by the local space.
  • the target confidence scores detected by all agents within the system are fused to generate the target. Specifically, it is expressed as follows:
  • the dynamic decision-making agent performs diffusion search and contraction search tasks based on global search information, specifically including: if All are 0, but there are intelligent agents. If True, meaning an agent has reached the search endpoint, then a spatial diffusion search is performed; if All values are 0, indicating no agent detected the target, but agents are present. If the value is True, it means an agent has reached the search endpoint. In this case, the search endpoints for all agents need to be updated, and the search range needs to be expanded, meaning the number of iterations ut increases. Because the search range expands, the local region... As this changes, it becomes necessary to inherit historical global information.
  • the inheritance formula is as follows:
  • the value is not 0, it indicates that an agent Si has detected the target and is performing a local agent space shrinkage search.
  • Corresponding local space As a suspected target space
  • the agents within eight local spaces around the target estimate the target location, thus achieving a shrinking search.
  • Step 5 of the present invention includes: based on the global search information of the intelligent agent, fusing various types of sensor data in the multi-agent system, and combining the gray wolf optimization algorithm to estimate the target location information, the perception information of the intelligent agent is as follows:
  • N is the number of agents; based on the agent's own state vector
  • the target detection probability is used to calculate the individual fitness value. Role assignment is completed based on the ranking of fitness values. The higher the target detection probability of an agent, the higher its fitness, indicating the emergence of a leader.
  • the fitness value calculation formula is as follows:
  • step 5 of the present invention the capture area is divided based on a dynamic Voronoi diagram, and the capture position of the agent is accurately determined by the estimated target coordinates (x goal , y goal ) and the coordinates of nearby obstacles. Including the coordinates of the alpha wolf, there are a total of seven coordinates.
  • Initializing the Voronoi diagram yields seven discrete regions. and and the vertex set of each region and
  • V be the area of the discrete Voronoi diagram region V_goal .
  • the wolf detection position update formula in the gray wolf optimization algorithm is improved to generate the updated wolf detection position.
  • the improved formula is as follows:
  • r1 and r2 are random vectors with values between [0,1], and a is a control parameter that decreases linearly from 2 to 0 during the iteration process.
  • Step 6 of the present invention includes: based on the real-time position of the agent and the position of the obstacle, if an agent detects a target, updating the agent's role using the Gray Wolf Optimization Algorithm, and updating the coordinates of obstacles near the leader wolf. and and target coordinates Generate a new Voronoi diagram; if no agent detects the target, but the target is moving, update the agent's decision and the coordinates of the obstacle after a time T of movement. and and target coordinates Generate a new Voronoi diagram; simultaneously, detect the area and boundary of the target object's Voronoi diagram, if the area... Less than the threshold If all boundaries are adjacent to Voronoi diagrams, then the encirclement ends.
  • a Voronoi diagram of the end of the encirclement is shown in Figure 4.

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Evolutionary Computation (AREA)
  • Artificial Intelligence (AREA)
  • Software Systems (AREA)
  • Computing Systems (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Mathematical Physics (AREA)
  • Multimedia (AREA)
  • Data Mining & Analysis (AREA)
  • Computational Linguistics (AREA)
  • General Engineering & Computer Science (AREA)
  • Health & Medical Sciences (AREA)
  • General Health & Medical Sciences (AREA)
  • Biophysics (AREA)
  • Molecular Biology (AREA)
  • Biomedical Technology (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Databases & Information Systems (AREA)
  • Medical Informatics (AREA)
  • Control Of Position, Course, Altitude, Or Attitude Of Moving Bodies (AREA)
  • Radar Systems Or Details Thereof (AREA)

Abstract

本发明公开了分布式启发的无人集群类脑群智融合搜索围捕方法，该方法包括：步骤1：基于Yolov5网络完成单体目标检测，并结合DeepSort算法，预测出下一帧目标出现的位置；步骤2：根据距离差和面积差映射出智能体在物理空间的导航终点坐标式（1）；步骤3：构建扩散收缩协调的混沌映射算法，将导航终点坐标的更新映射到分布均匀的混沌空间；步骤4：根据所有智能体的目标检测置信度决策信息与区域位置信息，融合生成搜索空间式（2）全局搜索信息式（3）；步骤5：根据智能体全局搜索信息，融合多智能体系统中的各类传感数据，估算目标位置信息；步骤6：根据智能体的实时位置和障碍物位置。本发明围捕效率高，能够精确更新智能体的围捕终点，完成对运动目标的包围任务。

Description

分布式启发的无人集群类脑群智融合搜索围捕方法 技术领域
本发明涉及分布式启发的无人集群类脑群智融合搜索围捕方法，尤其涉及利用启发式优化算法进行群体智能决策技术，属于人工智能与群体智能技术领域。
背景技术
在多智能体搜索技术的研究方面，国内外的研究主要针对目标区域的预先划分，设定智能体的搜索航线，从而完成多智能体的协同搜索任务。但搜索任务是动态且随机的，而在无人作战场景下，环境信息和地图信息均未知，无法完成搜索航线的预先设定，只能根据搜索任务过程中的实时信息进行实时决策。在多智能体围捕技术的研究方面，国外研究主要集中在控制算法的优化上。例如，基于深度强化学习的控制策略在提高智能体协同围捕效率方面取得了显著成果。国内研究主要集中在智能体路径规划算法的提升上。基于遗传算法的路径优化在解决搜索围捕中的路径规划问题上表现出色，遗传算法通过模拟生物进化过程中的自然选择和遗传机制，能够有效地搜索到最优路径，提高搜索围捕的效率和成功率，此外，国内研究还关注智能体之间的协同控制策略，如协同规划和协同执行，以提高整体搜索围捕的协同效率。
基于分布式启发的群智融合技术是当前多智能体进行协同搜索围捕的主要手段之一。但在无人作战场景中，存在覆盖范围广、搜索围捕难度大、目标位置定位难等特点，同时在无人作战场景中，由于智能体机动能力强，容易产生运动模糊等问题，造成目标检测效率低。
发明内容
本发明目的在于针对上述现有技术的缺陷和不足，提出了分布式启发的无人集群类脑群智融合搜索围捕方法，首先通过基于YOLOv5s的目标检测技术，针对目标实时运动的问题，利用预测的轨迹完成对动态目标的跟踪任务；其次，针对无人作战场景下搜索面积广的问题，基于区域扩散收缩原理，改进混沌映射算法，以扩散收缩协调的方式映射无人智能体的搜索终点，并利用决策级类脑融合机制，完成多智能体的决策融合，减少区域重复搜索；最后，针对运动目标围捕效率低的问题，基于灰狼优化算法的角色更新机制，结合动态的Voronoi区域划分，提出多智能体信息级类脑融合围捕机制，精确更新智能体的围捕终点，完成对运动目标的包围任务。
本发明为解决其技术问题所采用的技术方案是：分布式启发的无人集群类脑群智融合搜索围捕方法，所述该方法包括如下步骤：
步骤1：基于Yolov5网络完成单体目标检测，并结合DeepSort算法，预测出下一帧目标可能出现的位置，计算预测框中点与中轴的归一化距离差diffdet和预测框面积与限定的成像面积间的归一化面积差diffarea。
步骤2：根据距离差和面积差映射出智能体在物理空间的导航终点坐标
步骤3：构建扩散收缩协调的混沌映射算法，将导航终点坐标的更新映射到分布均匀的混沌空间。
步骤4：根据所有智能体的目标检测置信度决策信息与区域位置信息，融合生成搜索空间的全局搜索信息并根据全局搜索信息，动态决策智能体执行扩散搜索与收缩搜索任务。
步骤5：根据智能体全局搜索信息，融合多智能体系统中的各类传感数据，估计目标位置信息，基于动态的Voronoi图划分围捕区域，精确智能体的围捕位置。
步骤6：根据智能体的实时位置和障碍物位置，同时根据Voronoi图中目标物检测目标物Vgoal图的面积和边界，若面积小于阈值并且边界均有Voronoi图相邻，则结束围捕。
进一步地，所述步骤1中，基于Yolov5网络完成单体目标检测，并结合DeepSort算法，预测出下一帧目标可能出现的位置具体包括：预测的目标位置(x'det,y'det,width'det,height'det)如下：
(x'det,y'det,width'det,height'det)＝DeepSort(xdet,ydet,widthdet,heightdet)
其中，(xdet,ydet,widthdet,heightdet)为Yolov5目标检测框的中心点位置及检测框宽高；然后根据图像的宽度计算检测框中点与中轴的归一化距离差diffdet：
同时计算检测框的面积与限定的成像面积的归一化面积差：
其中，widthimg和heightimg分别为图像的宽和高。
进一步地，所述步骤2中，根据距离差和面积差映射出智能体在物理空间的导航终点坐标具体包括：首先根据归一化的距离差和面积差，估计智能体需要前进的距离和左右行进的距离：
其中，和分别为前进和左右行进的最大距离，Δ为最小实数，即规定检测框面积大于最佳面积时，智能体停止前进和后退，同时保证后续角度计算的可行性；映射的智能体全局坐标系下的导航终点坐标具体表达式为：
其中，和为智能体实时坐标，为智能体在全局坐标系下的偏转角。
进一步地，所述步骤3中，构建扩散收缩协调的混沌映射算法，将导航终点坐标的更新映射到分布均匀的混沌空间具体包括：首先将Mspace×Mspace的全局空间划分成m×m个局部空间当迭代ut次时，则将utMspace×utMspace的全局空间划分为utm×utm个局部空间然后以起始点为中心，以向外扩散的方式进行搜索，因此需利用改进的混沌映射估算出每个智能体搜索的目标位置，改进的混沌扩散因子表达式如下：
其中，μt为控制参数，其值范围为(0,1)，为混沌映射序列，其值范围为(0,1)，在Mspace×Mspace搜索空间内进行智能体初始搜索终点的混沌映射，其中原点位置位于Mspace×Mspace空间的中点，智能体Si搜索的目标位置估算表达式为：
其中，和分别为智能体搜索的终点坐标，ut为混沌映射迭代次数，生成初始搜索终点时ut的值为1，和分别为x轴和y轴的混沌映射因子。
进一步地，所述步骤3包括：若有智能体搜索到目标，即存在疑似目标空间，则在局部空间内进行混沌映射，将范围均为[-1,1]的x轴和y轴混沌因子映射到疑似目标空间的混沌因子映射范围内，混沌收缩因子的映射范围为：
混沌收缩因子表达式为：
其中，w和h为疑似目标空间的维度，计算公式分别如下：
其中，为x轴和y轴智能体Si的位置；此时利用混沌收缩因子，在疑似目标空间内进行目标位置估算，目标位置估算公式如下：
进一步地，所述步骤4中，根据所有智能体的目标检测置信度决策信息与区域位置信息，融合生成搜索空间的全局搜索信息具体包括：其中，智能体的决策信息包含如下：
其中，为智能体的目标检测置信度，为智能体Si是否到达导航终点，为智能体Si在搜索空间的区域位置；全局搜索信息具体表示如下：
其中，Cwh表示局部空间内检测到目标的智能体个数，Bwh表示搜索过局部空间的智能体个数；为局部空间内的目标检测置信度，由局部空间内所有的智能体检测到的目标置信度融合生成，具体表示为：
进一步地，所述步骤4中，根据全局搜索信息，动态决策智能体执行扩散搜索与收缩搜索任务具体包括：若均为0，但存在智能体的为True，即有智能体到达搜索终点，则进行空间扩散搜索，若均为0，没有智能体检测到目标，但存在智能体的为True，即有智能体到达搜索终点，此时需要对所有智能体的搜索终点进行更新，并且扩大搜索范围，即迭代步数ut增加，此时因为搜索范围扩大，局部区域也随之变化，需要对历史全局信息进行继承，继承公式如下：
若存在不为0，则表示有智能体Si检测到目标，进行局部智能体空间收缩搜索，则对应的局部空间作为疑似目标空间对局部空间周围八个局部空间内的智能体进行目标位置估算，实现收缩搜索。
进一步，所述步骤5中，根据智能体全局搜索信息，融合多智能体系统中的各类传感数据，结合灰狼优化算法估算目标位置信息具体包括：智能体的感知信息如下：
其中包含了激光雷达数据实时线速度和角速度已经实时的偏转角结合目标检测的状态信息能够组合成智能体Si的融合数据
同时接收到其他智能体的状态信息将其与自身的状态信息组合为全局状态信息数组，具体表示为：
其中，N为智能体的个数；基于智能体自身状态向量中的目标检测概率进行个体适应度值的计算，根据适应度值的排序结果完成角色分配，当智能体的目标检测概率越大，适应度就越高，即表示领头狼产生，适应度值计算公式表示为：
根据适应度值进行排序，获得领头狼Sα、Sβ、Sδ，并估算出正前方目标点的位置：
其中，为选择激光雷达正前方角度，以该角度内距离最近的障碍物坐标为目标物的坐标，为正前方角度中障碍物的距离，获取领头狼Sα、Sβ、Sδ各自正前方目标点 和并融合成精确的的目标点位置，具体表示如下：
根据激光雷达非正前方的剩余角度中距离最小的角度，计算Sα、Sβ、Sδ各自附近最近的障碍物坐标
其中，和分别为智能体Si的激光雷达数据中非正前方的剩余角度中距离最小的角度和距离。
进一步地，所述步骤5中，基于动态的Voronoi图划分围捕区域，精确智能体的围捕位置具体包括：由估计的目标物坐标(xgoal,ygoal)和其附近障碍物坐标以及领头狼的坐标，共七个坐标，完成Voronoi图的初始化可以得到七个离散区域和以及每个区域的顶点集合和计算目标物Vgoal图中的重心重心的计算公式如下：
其中，为离散Voronoi图区域Vgoal的面积，然后利用目标区域的重心位置对灰狼优化算法中的探狼位置更新公式进行改进，生成探狼的更新位置改进公式如下：
其中，A1和C1为协同系数：

C1＝2r2
其中，r1和r2为随机向量，其值在[0,1]之间，a是在迭代过程中从2线性递减至0的控制参数。
进一步地，所述步骤6包括：根据智能体的实时位置和障碍物位置，若有智能体检测到目标，利用灰狼优化算法更新智能体角色，并更新领头狼附近障碍物的坐标和和目标物坐标生成新的Voronoi图；若没有智能体检测到目标，但由于目标物是运动的，在运动T时间后，更新智能体决策，并更新障碍物的坐标和和目标物坐标生成新的Voronoi图；同时，检测目标物Vgoal图的面积和边界，若面积小于阈值并且边界均有Voronoi图相邻，则结束围捕。
有益效果：
1、本发明基于YOLOv5s的目标检测技术，针对目标实时运动的问题，利用预测的轨迹完成对动态目标的跟踪任务。
2、本发明针对无人作战场景下搜索面积广的问题，基于区域扩散收缩原理，改进混沌映射算法，以扩散收缩协调的方式映射无人智能体的搜索终点，并利用决策级类脑融合机制，完成多智能体的决策融合，减少区域重复搜索。
3、本发明针对运动目标围捕效率低的问题，基于灰狼优化算法的角色更新机制，结合动态的Voronoi区域划分，提出多智能体信息级类脑融合围捕机制，精确更新智能体的围捕终点，完成对运动目标的包围任务。
附图说明
图1为本发明分布式启发的无人集群类脑群智融合搜索围捕方法流程图。
图2为本发明目标检测框与图像的布局图。
图3为本发明迭代次数增加的搜索范围图。
图4为本发明围捕结束标志图。
具体实施方式
下面结合说明书附图对本发明创造做进一步地详细说明。
如图1所示，本发明提出了分布式启发的无人集群类脑群智融合搜索围捕方法，该方法包括如下步骤：
步骤1：基于Yolov5网络完成单体目标检测，并结合DeepSort算法，预测出下一帧目标出现的位置，计算预测框中点与中轴的归一化距离差diffdet和预测框面积与限定的成像面积间的归一化面积差diffarea；
步骤2：根据距离差和面积差映射出智能体在物理空间的导航终点坐标
步骤3：构建扩散收缩协调的混沌映射算法，将导航终点坐标的更新映射到分布均匀的混沌空间；
步骤4：根据所有智能体的目标检测置信度决策信息与区域位置信息，融合生成搜索空间的全局搜索信息并根据全局搜索信息，动态决策智能体执行扩散搜索与收缩搜索任务；
步骤5：根据智能体全局搜索信息，融合多智能体系统中的各类传感数据，估算目标位置信息，基于动态的Voronoi图划分围捕区域，精确智能体的围捕位置；
步骤6：根据智能体的实时位置和障碍物位置，同时根据Voronoi图中目标物检测目标物Vgoal图的面积和边界，若面积小于阈值并且边界均有Voronoi图相邻，则结束围捕。
本发明该方法具体操作步骤如下：
本发明上述步骤1包括：基于Yolov5网络完成单体目标检测，并结合DeepSort算法，预测出下一帧目标可能出现的位置，预测的目标位置(x'det,y'det,width'det,height'det)如下：
(x'det,y'det,width'det,height'det)＝DeepSort(xdet,ydet,widthdet,heightdet)
其中，(xdet,ydet,widthdet,heightdet)为Yolov5目标检测框的中心点位置及检测框宽高；然后根据图像的宽度计算检测框中点与中轴的归一化距离差diffdet：
同时计算检测框的面积与限定的成像面积的归一化面积差：
其中，widthimg和heightimg分别为图像的宽和高，检测框与图像的布局如图2所示。
本发明上述步骤2包括：基于Yolov5网络完成单体目标检测，根据距离差和面积差映射出智能体在物理空间的导航终点坐标首先根据归一化的距离差和面积差，估计智能体需要前进的距离和左右行进的距离：
其中，和分别为前进和左右行进的最大距离，Δ为最小实数，即规定检测框面积大于最佳面积时，智能体停止前进和后退，同时保证后续角度计算的可行性；映射的智能体全局坐标系下的导航终点坐标具体表达式为：
其中，和为智能体实时坐标，为智能体在全局坐标系下的偏转角。
本发明上述步骤3包括：构建扩散收缩协调的混沌映射算法，将导航终点坐标的更新映射到分布均匀的混沌空间，首先将Mspace×Mspace的全局空间划分成m×m个局部空间当迭代ut次时，则将utMspace×utMspace的全局空间划分为utm×utm个局部空间迭代更新的空间如图3所示，然后以起始点为中心，以向外扩散的方式进行搜索，因此需利用改进的混沌映射估计每个智能体搜索的目标位置，改进的混沌扩散因子表达式如下：
其中，μt为控制参数，其值范围为(0,1)，为混沌映射序列，其值范围为(0,1)，在Mspace×Mspace搜索空间内进行智能体初始搜索终点的混沌映射，其中原点位置位于Mspace×Mspace空间的中点，智能体Si搜索的目标位置估算表达式为：
其中，和分别为x轴和y轴智能体搜索的终点坐标，ut为混沌映射迭代次数，生成初始搜索终点时ut的值为1，和分别为x轴和y轴的混沌映射因子。
本发明上述步骤3中若有智能体搜索到目标，即存在疑似目标空间，则在局部空间内进行混沌映射，将范围均为[-1,1]的x轴和y轴混沌因子映射到疑似目标空间的混沌因子映射范围内，混沌收缩因子的映射范围为：

混沌收缩因子表达式为：
其中，w和h为疑似目标空间的维度，计算公式分别表示如下：
其中，为x轴和y轴智能体Si的位置；此时利用混沌收缩因子，在疑似目标空间内进行目标位置估计，目标位置估算公式如下：
本发明上述步骤4包括：根据所有智能体的目标检测置信度决策信息与区域位置信息，融合生成搜索空间的全局搜索信息其中，智能体的决策信息包含如下：
其中，为智能体的目标检测置信度，为智能体Si是否到达导航终点，为智能体Si在搜索空间的区域位置；全局搜索信息具体表示如下：
其中，Cwh表示局部空间内检测到目标的智能体个数，Bwh表示搜索过局部空间的智能体个数；为局部空间内的目标检测置信度，由局部空间内所有的智能体检测到的目标置信度融合生成，具体表示为：
本发明上述步骤4中，根据全局搜索信息，动态决策智能体执行扩散搜索与收缩搜索任务具体包括：若均为0，但存在智能体的为True，即有智能体到达搜索终点，则进行空间扩散搜索；若均为0，没有智能体检测到目标，但存在智能体的为True，即有智能体到达搜索终点，此时需要对所有智能体的搜索终点进行更新，并且扩大搜索范围，即迭代步数ut增加，此时因为搜索范围扩大，局部区域也随之变化，需要对历史全局信息进行继承，继承公式如下：
若存在不为0，则表示有智能体Si检测到目标，进行局部智能体空间收缩搜索，则对应的局部空间作为疑似目标空间对局部空间周围八个局部空间内的智能体进行目标位置估算，实现收缩搜索。
本发明上述步骤5包括：根据智能体全局搜索信息，融合多智能体系统中的各类传感数据，结合灰狼优化算法估算目标位置信息，智能体的感知信息如下：
其中包含了激光雷达数据实时线速度和角速度已经实时的偏转角结合目标检测的状态信息能够组合成智能体Si的融合数据
同时接收到其他智能体的状态信息将其与自身的状态信息组合为全局状态信息数组，具体表示为：
其中，N为智能体的个数；基于智能体自身状态向量中的目标检测概率进行个体适应度值的计算，根据适应度值的排序结果完成角色分配，当智能体的目标检测概率越大，适应度就越高，即表示领头狼产生，适应度值计算公式表示为：
根据适应度值进行排序，获得领头狼Sɑ、Sβ、Sδ，并估算出正前方目标点的位置：
其中，为选择激光雷达正前方角度，以该角度内距离最近的障碍物坐标为目标物的坐标，为正前方角度中障碍物的距离，获取领头狼Sɑ、Sβ、Sδ各自正前方目标点 和并融合成精确的的目标点位置，具体表示如下：
根据激光雷达非正前方的剩余角度中距离最小的角度，计算Sɑ、Sβ、Sδ各自附近最近的障碍物坐标
其中，和分别为智能体Si的激光雷达数据中非正前方的剩余角度中距离最小的角度和距离。
本发明上述步骤5中，基于动态的Voronoi图划分围捕区域，精确智能体的围捕位置，由估计的目标物坐标(xgoal,ygoal)和其附近障碍物坐标以及领头狼的坐标，共七个坐标，完成Voronoi图的初始化可以得到七个离散区域和以及每个区域的顶点集合和
计算目标物Vgoal图中的重心重心的计算公式如下：
其中，为离散Voronoi图区域Vgoal的面积。然后利用目标区域的重心位置对灰狼优化算法中的探狼位置更新公式进行改进，生成探狼的更新位置改进公式如下：
其中，A1和C1为协同系数：

C1＝2r2
其中，r1和r2为随机向量，其值在[0,1]之间，a是在迭代过程中从2线性递减至0的控制参数。
本发明上述步骤6包括：根据智能体的实时位置和障碍物位置，若有智能体检测到目标，利用灰狼优化算法更新智能体角色，并更新领头狼附近障碍物的坐标和和目标物坐标生成新的Voronoi图；若没有智能体检测到目标，但由于目标物是运动的，在运动T时间后，更新智能体决策，并更新障碍物的坐标和和目标物坐标生成新的Voronoi图；同时，检测目标物Vgoal图的面积和边界，若面积小于阈值并且边界均有Voronoi图相邻，则结束围捕，围捕结束的Voronoi示意图如图4所示。
应当指出，以上实施例的说明只是用于帮助理解本申请的方法及其核心思想，对于本技术领域的普通技术人员来说，在不脱离本申请原理的前提下，还可以对本申请进行若干改进和修饰，这些改进和修饰也在本申请权利要求的保护范围内。

Claims (10)

  1. 分布式启发的无人集群类脑群智融合搜索围捕方法，其特征在于，所述该方法包括如下步骤：
    步骤1：基于Yolov5网络完成单体目标检测，并结合DeepSort算法，预测出下一帧目标出现的位置，计算预测框中点与中轴的归一化距离差diffdet和预测框面积与限定的成像面积间的归一化面积差diffarea；
    步骤2：根据距离差和面积差映射出智能体在物理空间的导航终点坐标
    步骤3：构建扩散收缩协调的混沌映射算法，将导航终点坐标的更新映射到分布均匀的混沌空间；
    步骤4：根据所有智能体的目标检测置信度决策信息与区域位置信息，融合生成搜索空间的全局搜索信息并根据全局搜索信息，动态决策智能体执行扩散搜索与收缩搜索任务；
    步骤5：根据智能体全局搜索信息，融合多智能体系统中的各类传感数据，估算目标位置信息，基于动态的Voronoi图划分围捕区域，精确智能体的围捕位置；
    步骤6：根据智能体的实时位置和障碍物位置，同时根据Voronoi图中目标物检测目标物Vgoal图的面积和边界，若面积小于阈值并且边界均有Voronoi图相邻，则结束围捕。
  2. 根据权利要求1所述的分布式启发的无人集群类脑群智融合搜索围捕方法，其特征在于，所述步骤1中，基于Yolov5网络完成单体目标检测，并结合DeepSort算法预测出下一帧目标可能出现的位置具体包括：预测的目标位置(x'det,y'det,width'det,height'det)如下：
    (x'det,y'det,width'det,height'det)＝DeepSort(xdet,ydet,widthdet,heightdet)
    其中，(xdet,ydet,widthdet,heightdet)为Yolov5目标检测框的中心点位置及检测框宽高；然后根据图像的宽度计算检测框中点与中轴的归一化距离差diffdet：
    同时计算检测框的面积与限定的成像面积的归一化面积差：
    其中，widthimg和heightimg分别为图像的宽和高。
  3. 根据权利要求1所述的分布式启发的无人集群类脑群智融合搜索围捕方法，其特征在于，所述步骤2中，根据距离差和面积差映射出智能体在物理空间的导航终点坐标具体包括：首先根据归一化的距离差和面积差估算智能体需要前进的距离和左右行进的距离：
    其中，和分别为前进和左右行进的最大距离，Δ为最小实数，即规定检测框面积大于最佳面积时，智能体停止前进和后退，同时保证后续角度计算的可行性；映射的智能体全局坐标系下的导航终点坐标具体表达式为：
    其中，和为智能体实时坐标，为智能体在全局坐标系下的偏转角。
  4. 根据权利要求1所述的分布式启发的无人集群类脑群智融合搜索围捕方法，其特征在于，所述步骤3中，构建扩散收缩协调的混沌映射算法，将导航终点坐标的更新映射到分布均匀的混沌空间具体包括：首先将Mspace×Mspace的全局空间划分成m×m个局部空间当迭代ut次时，则将utMspace×utMspace的全局空间划分为utm×utm个局部空间然后以起始点为中心，以向外扩散的方式进行搜索，因此需利用改进的混沌映射估算出每个智能体搜索的目标位置，改进的混沌扩散因子表达式如下：
    其中，μt为控制参数，其值范围为(0,1)，为混沌映射序列，其值范围为(0,1)，在Mspace×Mspace搜索空间内进行智能体初始搜索终点的混沌映射，其中原点位置位于Mspace×Mspace空间的中点，智能体Si搜索的目标位置估算表达式为：
    其中，和分别为x轴和y轴智能体搜索的终点坐标，ut为混沌映射迭代次数，生成初始搜索终点时ut的值为1，和分别为x轴和y轴的混沌映射因子。
  5. 根据权利要求1所述的分布式启发的无人集群类脑群智融合搜索围捕方法，其特征在于，所述步骤3包括：若有智能体搜索到目标，即存在疑似目标空间，则在局部空间内进行混沌映射，将范围均为[-1,1]的x轴和y轴混沌因子映射到疑似目标空间的混沌因子映射范围内，混沌收缩因子的映射范围为：
    混沌收缩因子表达式为：
    其中，w和h为疑似目标空间的维度，计算公式分别表示如下：
    其中，为x轴和y轴智能体Si的位置；此时利用混沌收缩因子，在疑似目标空间内进行目标位置估算，目标位置估算公式如下：
  6. 根据权利要求1所述的分布式启发的无人集群类脑群智融合搜索围捕方法，其特征在于，所述步骤4中，根据所有智能体的目标检测置信度决策信息与区域位置信息，融合生成搜索空间的全局搜索信息具体包括：其中，智能体的决策信息包含如下：
    其中，为智能体的目标检测置信度，为智能体Si是否到达导航终点，为智能体Si在搜索空间的区域位置；全局搜索信息具体表示如下：
    其中，Cwh表示局部空间内检测到目标的智能体个数，Bwh表示搜索过局部空间的智能体个数；为局部空间内的目标检测置信度，由局部空间内所有的智能体检测到的目标置信度融合生成，具体表示为：
  7. 根据权利要求1所述的分布式启发的无人集群类脑群智融合搜索围捕方法，其特征在于，所述步骤4中，根据全局搜索信息，动态决策智能体执行扩散搜索与收缩搜索任务具体包括：若均为0，但存在智能体的为True，即有智能体到达搜索终点，则进行空间扩散搜索；若均为0，没有智能体检测到目标，但存在智能体的为True，即有智能体到达搜索终点，此时需要对所有智能体的搜索终点进行更新，并且扩大搜索范围，即迭代步数ut增加，此时因为搜索范围扩大，局部区域也随之变化，需要对历史全局信息进行继承，继承公式如下：
    若存在不为0，则表示有智能体Si检测到目标，进行局部智能体空间收缩搜索，则对应的局部空间作为疑似目标空间对局部空间周围八个局部空间内的智能体进行目标位置估算，实现收缩搜索。
  8. 根据权利要求1所述的分布式启发的无人集群类脑群智融合搜索围捕方法，其特征在于，所述步骤5中，根据智能体全局搜索信息，融合多智能体系统中的各类传感数据，结合灰狼优化算法估算目标位置信息具体包括：智能体的感知信息如下：
    其中包含了激光雷达数据实时线速度和角速度已经实时的偏转角结合目标检测的状态信息能够组合成智能体Si的融合数据
    同时接收到其他智能体的状态信息将其与自身的状态信息组合为全局状态信息数组，具体表示为：
    其中，N为智能体的个数；基于智能体自身状态向量中的目标检测概率进行个体适应度值的计算，根据适应度值的排序结果完成角色分配，当智能体的目标检测概率越大，适应度就越高，即表示领头狼产生，适应度值计算公式表示为：
    根据适应度值进行排序，获得领头狼Sα、Sβ、Sδ，并估算出正前方目标点的位置：
    其中，为选择激光雷达正前方角度，以该角度内距离最近的障碍物坐标为目标物的坐标，为正前方角度中障碍物的距离，获取领头狼Sα、Sβ、Sδ各自正前方目标点 和并融合成精确的的目标点位置Xgoal，具体表示如下：
    根据激光雷达非正前方的剩余角度中距离最小的角度，计算Sα、Sβ、Sβ各自附近最近的障碍物坐标
    其中，和分别为智能体Si的激光雷达数据中非正前方的剩余角度中距离最小的角度和距离。
  9. 根据权利要求8所述的分布式启发的无人集群类脑群智融合搜索围捕方法，其特征在于，所述步骤5中，基于动态的Voronoi图划分围捕区域，精确智能体的围捕位置具体包括：由估计的目标物坐标(xgoal,ygoal)和其附近障碍物坐标以及领头狼的坐标，共七个坐标，完成Voronoi图的初始化，能够得到七个离散区域和以及每个区域的顶点集合和
    计算目标物Vgoal图中的重心重心的计算公式如下：
    其中，为离散Voronoi图区域Vgoal的面积，然后利用目标区域的重心位置对灰狼优化算法中的探狼位置更新公式进行改进，生成探狼的更新位置改进公式如下：
    其中，A1和C1为协同系数：

    C1＝2r2
    其中，r1和r2为随机向量，其值在[0,1]之间，a是在迭代过程中从2线性递减至0的控制参数。
  10. 根据权利要求10所述的分布式启发的无人集群类脑群智融合搜索围捕方法，其特征在于，所述步骤6包括：根据智能体的实时位置和障碍物位置，若有智能体检测到目标，利用灰狼优化算法更新智能体角色，并更新领头狼附近障碍物的坐标和和目标物坐标生成新的Voronoi图；若没有智能体检测到目标，但由于目标物是运动的，在运动T时间后，更新智能体决策，并更新障碍物的坐标和和目标物坐标生成新的Voronoi图；同时，检测目标物Vgoal图的面积和边界，若面积小于阈值并且边界均有Voronoi图相邻，则结束围捕。
PCT/CN2024/136298 2024-05-10 2024-12-03 分布式启发的无人集群类脑群智融合搜索围捕方法 Pending WO2025232168A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN202410574822.4 2024-05-10
CN202410574822.4A CN118608559B (zh) 2024-05-10 2024-05-10 分布式启发的无人集群类脑群智融合搜索围捕方法

Publications (1)

Publication Number Publication Date
WO2025232168A1 true WO2025232168A1 (zh) 2025-11-13

Family

ID=92548754

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2024/136298 Pending WO2025232168A1 (zh) 2024-05-10 2024-12-03 分布式启发的无人集群类脑群智融合搜索围捕方法

Country Status (2)

Country Link
CN (1) CN118608559B (zh)
WO (1) WO2025232168A1 (zh)

Families Citing this family (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN118608559B (zh) * 2024-05-10 2025-10-03 南京邮电大学 分布式启发的无人集群类脑群智融合搜索围捕方法
CN121257641B (zh) * 2025-10-11 2026-03-20 华东师范大学 一种用于具身导航决策理解的智能体训练方法

Citations (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114973443A (zh) * 2022-05-19 2022-08-30 杭州中威电子股份有限公司 一种基于巡检机器人的复杂气体环境巡检系统及其方法
CN115019185A (zh) * 2022-08-03 2022-09-06 华中科技大学 类脑连续学习协同围捕方法、系统及介质
CN115311883A (zh) * 2022-08-09 2022-11-08 杭州中威电子股份有限公司 一种基于机器学习的智慧停车系统及其方法
CN115550837A (zh) * 2022-09-22 2022-12-30 合肥工业大学 一种基于混沌映射与灰狼算法优化的DV-Hop定位方法
US20230124356A1 (en) * 2020-09-18 2023-04-20 Wuhan University Of Technology Coverage-path planning method for single unmanned surface mapping vessel
CN116592886A (zh) * 2023-04-29 2023-08-15 西北工业大学深圳研究院 一种改进的混沌差分鲸鱼算法的智能体路径优化方法
CN118608559A (zh) * 2024-05-10 2024-09-06 南京邮电大学 分布式启发的无人集群类脑群智融合搜索围捕方法

Family Cites Families (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN116935058B (zh) * 2023-06-09 2025-11-04 南京邮电大学 基于分布式强化学习的无人集群协同群智目标搜索方法
CN117629210A (zh) * 2023-11-03 2024-03-01 清华大学 基于图网络的可拓展多智能体分层导航方法及装置

Patent Citations (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20230124356A1 (en) * 2020-09-18 2023-04-20 Wuhan University Of Technology Coverage-path planning method for single unmanned surface mapping vessel
CN114973443A (zh) * 2022-05-19 2022-08-30 杭州中威电子股份有限公司 一种基于巡检机器人的复杂气体环境巡检系统及其方法
CN115019185A (zh) * 2022-08-03 2022-09-06 华中科技大学 类脑连续学习协同围捕方法、系统及介质
CN115311883A (zh) * 2022-08-09 2022-11-08 杭州中威电子股份有限公司 一种基于机器学习的智慧停车系统及其方法
CN115550837A (zh) * 2022-09-22 2022-12-30 合肥工业大学 一种基于混沌映射与灰狼算法优化的DV-Hop定位方法
CN116592886A (zh) * 2023-04-29 2023-08-15 西北工业大学深圳研究院 一种改进的混沌差分鲸鱼算法的智能体路径优化方法
CN118608559A (zh) * 2024-05-10 2024-09-06 南京邮电大学 分布式启发的无人集群类脑群智融合搜索围捕方法

Also Published As

Publication number Publication date
CN118608559A (zh) 2024-09-06
CN118608559B (zh) 2025-10-03

Similar Documents

Publication Publication Date Title
CN114119920B (zh) 三维点云地图构建方法、系统
CN103455034B (zh) 一种基于最近距离向量场直方图的避障路径规划方法
CN114002667B (zh) 一种基于随机矩阵法的多近邻扩展目标跟踪算法
CN109341707A (zh) 未知环境下移动机器人三维地图构建方法
CN111610786A (zh) 基于改进rrt算法的移动机器人路径规划方法
CN118608559B (zh) 分布式启发的无人集群类脑群智融合搜索围捕方法
CN120595814A (zh) 一种机器人路径规划方法
CN106873599A (zh) 基于蚁群算法和极坐标变换的无人自行车路径规划方法
CN118092418A (zh) 一种多无人机协同跟踪地面移动目标的控制方法
CN115268448A (zh) 一种基于冲突搜索和速度障碍的多机器人路径规划方法
CN117075607B (zh) 一种适用于复杂环境的改进jps的无人车路径规划方法
CN119105524A (zh) 基于机器学习的无人机定位导航与路径规划优化方法
CN117389305A (zh) 一种无人机巡检路径规划方法、系统、设备及介质
CN117270562A (zh) 一种基于复杂离散环境下rrt和vo的单无人机避障路径规划方法
CN117970928A (zh) 基于ros系统多移动机器人协同的路径规划方法
CN120506970A (zh) 基于改进双向a*和dwa算法的室外机器人路径规划方法
CN120333437A (zh) 无人机多目标影像采集路径规划方法、设备及存储介质
Bu et al. RUSH: rapid UAV spatial hierarchical exploration via regional viewpoint generation for large-scale environments
CN120779970B (zh) 一种基于Astar算法的安全性增强轨迹规划方法
CN120370956B (zh) 基于高密度雷达点云的三维动态路径规划方法
Sari et al. Ant colony optimization-based path planning for autonomous vehicle navigation systems
CN119573706A (zh) 一种基于全局静态地图和局部动态地图的定位方法及应用
CN118857319A (zh) 一种粮食清扫车及其路径规划方法
CN112650238B (zh) 一种利用可见性信息的实时路径规划方法
Ye et al. Rgbd-based image goal navigation with pose drift: A topo-metric graph based approach

Legal Events

Date Code Title Description
ENP Entry into the national phase

Ref document number: 2025537946

Country of ref document: JP

Kind code of ref document: A

WWE Wipo information: entry into national phase

Ref document number: 2025537946

Country of ref document: JP

121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 24940508

Country of ref document: EP

Kind code of ref document: A1