| The popularity of intelligent mobile devices and the rapid development of the Internet have led to the emergence of a large number of new services.However,different types of services have different requirements for network resources,and resource managers need to make better resource allocation based on different service requirements.In this situation,this thesis first designs an adaptive service type identification method based on service diversity.In order to further meet the differentiated needs of diverse services,a highly reliable access strategy based on NOMA is designed.First of all,in view of the problems of low user privacy security,poor network adaptability,and lack of integrity in feature extraction in existing service type identification methods,this thesis proposes an adaptive service classification method based on a stacked hybrid auto-encoder,which achieves better classification results.Based on the needs of user privacy and the integrity of service features,the algorithm obtains the external features of service as the input of service type identification method.In order to improve network adaptability and obtain the feature extraction model which can meet the requirements of different service classification scenarios,a method of adjusting the number of hidden layer neurons in a small range is proposed to reduce training costs and improve adaptability.By comparing the distance between the feature extracted from the unknown service and the feature center of each type of service,the label of the unknown service is obtained.Applying the algorithm to the ISCXTor2016 dataset,the average accuracy rate of 85%and the average recall rate of 82%are obtained,both higher than other methods,which proves the effectiveness of the proposed method.Secondly,in order to address the issue of insufficient consideration of diverse service differentiation requirements when studying transmission reliability,this thesis focuses on the optimization of user transmission timeslots in the NOMA uplink multi-service transmission process.By scheduling time domain resources,the number of packets transmitted successfully is maximized to ensure system reliability.Considering the differentiated requirements of different services for different QoS elements,corresponding weighting factors are established for each type of service,and the weighting value of the number of packets transmitted successfully is taken as the optimization goal.Because the problem is non-convex and difficult to solve directly,a method based on multi-agent reinforcement learning is proposed.The simulation results show that the method performs well in multi-service scenarios,which can improve the number of packets transmitted successfully,thereby improving system reliability,and has lower latency,indicating the effectiveness and superiority of the proposed method. |
