Research On Resource Allocation Strategy Of Space TT&C Network Based On Automatic Machine Learning

The shortage of radio spectrum resources is becoming increasingly prominent,as is the case in the field of space TT&C networks.In order to solve the above problems,artificial intelligence technology has been introduced into the scenario of spectrum resource allocation in the aerospace measurement and control network.On the basis of summarizing previous research results,this thesis explores the spectrum allocation problem of aerospace measurement and control networks using automatic machine learning technology.This thesis first reviews the current research status of international aerospace measurement and control networks,classic excellent algorithms for spectrum allocation,and existing achievements in the application of artificial intelligence technology in the field of aerospace measurement and control.It proposes the ideas of automatic machine learning algorithms based on three neural architecture search strategies for spectrum recognition and automatic machine learning algorithms based on Bayesian optimization search strategies for spectrum resource allocation.Secondly,this thesis proposes a spectrum recognition method based on automatic machine learning algorithms.This thesis takes the power spectral density of the classic aerospace channel model as the data source,inputs them into the neural network after coding,and completes the task of spectrum recognition of the aerospace TT&C network communication model,and compares it with the slice cyclic neural network algorithm that is manually tuned and has good recognition effect.The simulation results show that the main evaluation indicators such as accuracy and loss rate when applying automatic machine learning algorithms to spectrum recognition are not significantly different from those of manually tuned neural networks,and have certain practical significance.Finally,this thesis proposes a spectrum allocation method for automatic machine learning algorithms based on Bayesian optimization search strategy.Through the previous stage of experiments,it was found that the spectrum recognition effect based on Bayesian optimization search strategy is the best,so it was applied to the prediction of network traffic.The data source still comes from the sampling of the power spectral density of the above classical channel model,adds interference to simulate the real environment,and inputs it into the generated neural network to predict network traffic.Then,the predicted results are used as input data to complete the task of spectrum allocation for the communication model of the aerospace TT&C network,and compared with the adaptive path idle degree algorithm manually adjusted.The simulation results show that applying automatic machine learning algorithms to spectrum allocation,the main evaluation indicators such as network resource utilization and network bandwidth blocking rate,although slightly inferior to the adaptive path idle degree algorithm,still have strong practicality due to its ability to automatically generate neural network architectures and reduce the task of manual parameter tuning.