Research On Multi-AUV Cooperative Hunting Method In Complex Underwater Environment

In recent years,with the rapid development of science and technology,human beings have become more and more concerned about marine resources.However,due to the special nature of the marine environment,it is difficult to achieve autonomous protection and effective management by human resources alone,which limits the exploration and development of the ocean.This has promoted the rapid development of AUV.Target detection and identification is the key to achieve successful target seizure.But the interference of other factors such as turbid water and insufficient light makes it difficult for AUVs to acquire effective characteristics of targets.The small number of underwater samples makes it difficult for the algorithm to accurately identify targets with insufficient training.The underwater communication delays and time-varying currents greatly affect the cooperative control of multi-AUV systems during the roundup process.This paper addresses the problems of limited information collection by a single AUV,interference in complex underwater environment and difficulty in obtaining effective data of target features,etc.,and uses multi-AUV collaboration mechanism and migration reinforcement learning model to detect the seized targets and increase the detection capability of the multi-AUV system for the seized targets.To address the problem of insufficient target training samples,the GAN-meta learning method is used to train the model and improve the recognition capability of the algorithm.To address the problems of underwater communication delay and time-varying current interference,the GAN model is used to automatically generate the appropriate cooperative control rate and combine with the multi-AUV topology to achieve the successful seizure of non-cooperative targets.The details of the above proposed approach are as follows.(1)A collaborative multi-AUV target detection method based on transfer-reinforcement learning is proposed.The collected target information is fused with features using wavelet transform and affine invariance,and the similarity of the features is calculated based on the Marxian distance.The learning model is selected autonomously based on the similarity threshold.When the similarity is small,the target information is reinforced and trained to reduce environmental interference.When the similarity is large,the source domain feature data are migrated to the target domain to reduce the repeated computation of similar data and ensure the real-time performance of the algorithm.(2)A GAN-meta learning based dangerous target recognition method is proposed.The VGG-19 network and WGAN network are used for feature extraction and information complementation of target images to reduce the influence of environmental interference on the acquired images.The meta-learning theory is invoked to train the parameter changes in the feature extraction process to improve the algorithm’s ability to recognize new targets and to ensure that the model has a strong generalization capability.(3)A GAN-based multi-AUV consistency cooperative control method is proposed.First,the 3D kinematic model of AUV is established.Then,the topology under ideal environment is established by combining Laplace matrix,and the control rate of AUV is calculated.Finally,the GAN network model is invoked,and the multi-AUV control rate adapted to the current disturbance environment is generated by iterative training of GAN.The impact of time-varying currents and hydroacoustic communication delays on the consistency of the multi-AUV system is reduced,and the success rate of target hunting is improved.(4)The proposed algorithm is validated by simulations and compared with existing methods,and the experimental results show that the proposed method has better underwater target detection and identification and seizure capabilities.