Research On Intelligent Effectiveness Evaluation Method Of UAV Swarm Based On CapsNet

With the continuous development of system simulation technology,the traditional effectiveness evaluation and analysis methods have been difficult to fully explore and excavate massive data of system simulation,resulting in inadequate evaluation and analysis conclusions.In addition,the traditional methods are difficult to support the search of design space for complex simulation experiment,so as to obtain the simulation scheme of effectiveness optimization,which greatly affects the role of effectiveness evaluation for the actual problems in system construction and operational application.An effective way to solve the above problems is to construct an effectiveness evaluation model based on deep learning with partial simulation data by means of artificial intelligence technology.However,applying deep learning to effectiveness evaluation will bring two problems: First,deep learning is difficult to process non-sequential one-dimensional simulation data.Second,deep learning always gets poor performance on small-scale datasets.Based on the background of UAV swarm operation,this paper studies the intelligent evaluation of system combat effectiveness,and solves the problem that conventional deep learning methods are difficult to use in the field of effectiveness evaluation.It mainly includes the following aspects:(1)The ABMS-based UAV swarm defense simulation system is established.The behavior characteristics and rules of the swarm combat between red and blue UAVs are studied.The process of UAV swarm operation is analyzed through simulation,and the reasonable indicator is proposed for evaluation,which will provide reasonable engineering design advice for similar defense systems in the future.(2)A new structure of capsule neural network based on dual-path module(DPCNet)is proposed,which further improves the classification accuracy by combining the advantages of dual-path neural network and CapsNet.The DPCNet achieves excellent results on two public datasets,improves the performance of the model in processing complex datasets,and reduces computational overhead.(3)A general framework for effectiveness evaluation based on DPCNet is proposed,which consists of two parts: first,a data cyclic stacking operation is proposed to realize the conversion of one-dimensional data into two-dimensional data,which helps deep learning models to mine more correlation information from simulation data;second,the DPCNet is applied to solve the problem of poor performance on small-scale datasets in deep learning methods.The framework provides a new theoretical method and technical support for future research on effectiveness evaluation issues.(4)The DPCNet-based effectiveness evaluation framework is applied to solve practical problems,and the effectiveness evaluation model is constructed for the simulation system of UAV swarm defensive operation.The DPCNet model is trained by taking the factor set of simulation experiment as the network input,and effectiveness evaluation values obtained by a large number of simulation as the output.Compared with the other seven algorithms,the DPCNet model is superior in the research of effectiveness evaluation.In order to further improve the accuracy,this paper proposes an effectiveness evaluation model based on hybrid voting mechanism,and uses the samples outside the dataset to verify the validity of the model again.Finally,the evaluation model is used to search the huge sample space,and a set of optimal schemes satisfying the requirements is obtained.Then we provide some reasonable suggestions for engineering design and tactics application of UAV swarm operation system.