| The research on cognitive sonar system is a forward-looking and challenging research topic.At present,military powers and relevant organizations have shown great interest in cognitive sonar.It is an inevitable trend to achieve the cognitive function of sonar system with reinforcement learning algorithm.It is of great practical significance to use reinforcement learning algorithm to track unmanned underwater vehicles.In this paper,we try to combine reinforcement learning algorithm with passive cognitive sonar system.In this paper,the origin and research trend of cognitive sonar are summarized.At the same time,the advantages and disadvantages of intelligent motion planning methods and their optimization are analyzed and listed when UUV system implements different tasks such as detection and obstacle avoidance.Finally,reinforcement learning algorithm is selected to achieve the cognitive function of sonar system.The theoretical analysis and simulation verification of bearings only tracking can be used as the basis for the motion planning of unmanned underwater vehicles.In order to complete the research of the cognitive sonar system motion planning based on the Deep Deterministic Policy Gradient algorithm,firstly,the principle and workflow of DDPG are studied.And the design of reward function,the selection of activation function and the construction of neural network model are analyzed.Then,using Python language to build the motion planning environment model on the Tensor Flow learning framework and simulate the relative motion situation when UUVs join and the learning situation of UUV system at this time.The results show that the proposed motion planning method of passive cognitive sonar system is effective and provides a new idea for the development of cognitive sonar. |
