Optimization of Resource Conflicts in Network Slicing Based on Adaptive Dynamic Prediction
Network Slicing(NS)is a key technology in 5G networks,playing an important role in multi service dynamic scenarios.To address the resource conflict problem caused by dynamic changes in slicing requirements in 5G network slicing,an optimization method based on Adaptive Dynamic Prediction(ADP)is adopted,and an optimization scheme called"adaptive dynamic prediction-model optimization"is proposed.In the adaptive dynamic prediction module,the fluctuation level of dynamic slice traffic is divided to ensure the accuracy and adaptability of slice traffic prediction.Based on the partitioning results,two different Recurrent Neural Network(RNN)algorithms are used to predict the future traffic demand of slices,including point prediction based on Attention mechanism-Bidirectional Gated Recurrent Unit(Att-BiGRU)and interval prediction based on bootstrap method BiGRU.In the model optimization module,the user satisfaction function and the cost of slicing optimization configuration are defined based on the predicted results,and the resource conflict optimization problem is represented as maximizing network benefits.Due to the possibility of uncertain parameters in the output of the prediction module,a slice optimization configuration scheme is solved based on robust optimization and particle swarm optimization algorithms with variable particle numbers.In the simulation section,the proposed optimization scheme is validated.This method not only meets the dynamic requirements of slicing,but also reduces the negative impact of resource conflicts.It outperforms the comparison algorithm in terms of network revenue and request acceptance rate,with a link resource utilization rate of over 90%.
国家科技期刊平台
NSTL
万方数据
维普
