Research On Cooperative Jamming Decision Based On Deep Reinforcement Learning

As a key component of electronic countermeasures,radar jamming has become an important means to effectively fight for electromagnetic power.In order to break through the performance limitations of single jamming equipment and effectively seize the initiative of electronic countermeasures,it is urgent to carry out research on collaborative jamming decision-making method.Aiming at the problem that the traditional cooperative jamming decision-making method relies on prior information,which makes it difficult to adapt to noncooperative dynamic changing scenarios,this thesis studies the cooperative jamming decisionmaking technology based on deep reinforcement learning around the active jamming pattern.Firstly,taking radar as the interference object,combing the related processing technology under different working conditions,combining signal processing and data processing,the modeling of search and tracking radar is realized.Secondly,the typical radar active jamming modes are studied,and their influence on radar search and detection is analyzed.By using the non-uniform repeater jamming technology,a variable pulse width dexterous jamming based on intermittent chaotic sampling is proposed,which avoids the strong regular defect of repeater deception jamming.Then,the multi-machine cooperative jamming strategy for radar search process is studied,and the action,state and reward elements are designed.Combined with the multi-agent deep reinforcement learning theory,a centralized cooperative decision training framework is built,and a cooperative jamming decision algorithm based on multi-agent deep deterministic strategy gradient is proposed.Finally,the multi-aircraft cooperative track deception jamming strategy for radar tracking process is studied,and the coupling relationship between the jammer and the false target is established.Combined with the idea of value decomposition and differential reward,the cooperative update method of network model is optimized,and a cooperative jamming decision algorithm based on the improved multi-agent deep deterministic strategy gradient is proposed to solve the problem of multi-agent reward allocation and improve the ability of cooperative jamming decision.Based on the above research,a complete collaborative jamming strategy plan was formed,and the simulation results showed that under the condition that the number of masking and escort jamming machines was 5,the success rate of this collaborative jamming strategy plan reached 82%,which improves the reliability of cooperative jamming decision-making in noncooperative dynamic change scenarios and provided new ideas for promoting the combination of deep reinforcement learning and radar collaborative jamming decision making.