| Mobile swarm intelligence sensing technology refers to the use of group mobile devices in the network to autonomously perceive the surrounding environment,so as to make information collection more flexible and efficient,which is a promising way of data acquisition.The amount,quality and acquisition speed of data collected by mobile devices reflect the efficiency of swarm intelligence task completion.In the current research work,the collected data usually exist in a static state,and the importance of the data tends to be consistent.When UAV completes a data acquisition task,it usually takes a fixed energy unit as the premise to obtain the maximum amount of data.Meanwhile,in the process of data acquisition,the efficiency of information exchange between the agent and the object to be collected is low.According to the above research content,the shortcomings of the current research are as follows: 1)there is a lack of research on the information difference and dynamic change of the collected objects;2)There is a lack of research on how to maximize data quality while avoiding collision between agents and obstacles;3)In the process of data acquisition,there is a lack of research on how to establish a stable optimal matching relationship between the agent and the object to be collected;4)In the field of data acquisition,there is a lack of research on realizing efficient collaborative cooperation through information exchange between agents.In view of the above problems,the main research content of this paper is as follows:1.To address problems 1 and 2,based on the "peer-to-peer collaboration approach" model among intelligences,this paper designs a multi-intelligence cooperative data collection method and proposes a value-driven peer-to-peer collaborative group intelligence perception algorithm.Firstly,through the design of environment perception,position perception and velocity perception models,we establish the perception between intelligent bodies and targets with different degrees of importance and different motion states,and solve the problem of collecting the target objects with different and dynamic characteristics.In addition,guided rewards such as intelligent body movement amplitude reward,intelligent body distance reward,data acquisition reward and collision avoidance reward are designed to construct algorithmic reward return functions to achieve collaborative cooperation and effective collision avoidance among intelligent bodies.It also incorporates the current mainstream multi-intelligent body reinforcement learning algorithm model MADDPG to find the optimal strategy by centralized training and decentralized execution.Through multiple types of experimental scenarios such as static and dynamic analysis and validation of the algorithm’s parameter combinations and experimental proofs,the experimental results show that the proposed DDCWR method outperforms classical algorithms such as MADDPG and DDPG in several evaluation metrics such as data quality and energy efficiency,with a 5-6 times improvement in data quality and at least 20% improvement in energy efficiency.This result demonstrates the superior performance of the DDCWR method.2.To address problems 3 and 4,based on the "master-slave collaboration approach" model between intelligences,this paper designs a master-slave collaboration scenario with a commander-drone as the research object,and proposes a master-slave collaborative group intelligence sensing algorithm based on matching game and communication mechanism.Firstly,by introducing the idea of Gale-shapely matching game algorithm,an optimal and stable matching between the energy attributes of UAV and the data quality attributes of the target to be collected is established to realize a collection strategy based on the priority of data importance.In addition,to ensure the UAV’s continuous and specific attention to high-quality targets,this paper combines the MAAC framework,which is currently a more popular communication rule-based multi-intelligence reinforcement learning algorithm,and introduces a multi-attention mechanism module to achieve efficient information exchange and sharing between master-slave intelligences in the data acquisition process.Experiments show that the proposed c-MGCM method outperforms classical algorithms such as MADDPG and DDPG in several evaluation metrics such as reward value and distance value of matched pairs,with a 2-3 times improvement in reward value and at least 14% improvement in data quality.The results demonstrate the efficiency and stability of the c-MGCM method. |
